def zoom_rot(ii,dd):
""" Rotate and zoom an image around a given angle"""
a = np.random.randint(-10,10)
ddr = rotate(dd,a, order=0, prefilter=False)
iir = rotate(ii.transpose((1,2,0)),a, order=0, prefilter=False)
f = np.random.randint(10000,15100) / 10000.
h = int(dd.shape[0] / f)
w = int(dd.shape[1] / f)
s_fh = float(dd.shape[0]) / float(h)
s_fw = float(dd.shape[1]) / float(w)
s_f = (s_fh + s_fw) / 2.
offset = 0
cy = np.random.randint(offset,dd.shape[0] - h - offset + 1)
cx = np.random.randint(offset,dd.shape[1] - w - offset + 1)
ddc = ddr[cy:cy+h, cx:cx+w]
iic = iir[cy:cy+h,cx:cx+w,:]
dd_s = zoom(ddc,(s_fh, s_fw),order=0, prefilter=False)
dd_s /= s_f
ii_s = iic.transpose((2,0,1))
ii_s = zoom(ii_s,(1,s_fh,s_fw),order=0, prefilter=False)
return ii_s.astype(np.float32), dd_s.astype(np.float32)
def setUpClass(cls):
img = imread(get_path('AS15-M-0298_SML.png'), flatten=True)
img_coord = (482.09783936, 652.40679932)
cls.template = sp.clip_roi(img, img_coord, 5)
cls.template = rotate(cls.template, 90)
cls.template = imresize(cls.template, 1.)
cls.search = sp.clip_roi(img, img_coord, 21)
cls.search = rotate(cls.search, 0)
cls.search = imresize(cls.search, 1.)
cls.offset = (1, 1)
cls.offset_template = sp.clip_roi(img, np.add(img_coord, cls.offset), 5)
cls.offset_template = rotate(cls.offset_template, 0)
cls.offset_template = imresize(cls.offset_template, 1.)
cls.search_center = [math.floor(cls.search.shape[0]/2),
math.floor(cls.search.shape[1]/2)]
cls.upsampling = 10
cls.alpha = math.pi/2
cls.cifi_thresh = 90
cls.rafi_thresh = 90
cls.tefi_thresh = 100
cls.use_percentile = True
cls.radii = list(range(1, 3))
cls.cifi_number_of_warnings = 2
cls.rafi_number_of_warnings = 2
def outputShiftedImage(self):
'''
Outputs a FITS file in which the slits have been shifted
to the best fitted positions.
'''
outfile1 = 'fittedSlitImage.fits'
outfile2 = 'fittedSlitImageFullFrame.fits'
zeros = np.zeros(self.slitImage.shape)
r = []
for res in self.result['minimaPosition'].values():
r.append(res[0])
x = res[1]
y = res[2]
n = res[5]
d = self.slits[n]['values']
xmin = self.slits[n]['xminSky'] + x
xmax = self.slits[n]['xmaxSky'] + x
ymin = self.slits[n]['yminSky'] + y
ymax = self.slits[n]['ymaxSky'] + y
zeros[ymin:ymax + 1, xmin:xmax + 1] = d
rot = np.median(np.asarray(r))
#note: -rot, because when fitting the direct image was rotated not the slits
img = interpolation.rotate(zeros, -rot, reshape=False)
if self.debug:
print '\n{0:.2f} degree rotation to the fits file'.format(-rot)
#output to a fits file
hdu = PF.PrimaryHDU(img)
if os.path.isfile(outfile1):
os.remove(outfile1)
hdu.writeto(outfile1)
#output a second image
zeros = np.zeros((3096, 3096))
for slit in self.slits:
xmin = self.slits[slit]['xminFitted']
xmax = self.slits[slit]['xmaxFitted']
ymin = self.slits[slit]['yminFitted']
ymax = self.slits[slit]['ymaxFitted']
zeros[ymin - self.slits[slit]['tolerance']:ymax + 1 + self.slits[slit]['tolerance'],\
xmin - self.slits[slit]['tolerance']:xmax + 1 + self.slits[slit]['tolerance']] = \
self.slits[slit]['valuesLarge']
#note: -rot, because when fitting the direct image was rotated not the slits
img = interpolation.rotate(zeros, -rot, reshape=False)
#output to a fits file
hdu = PF.PrimaryHDU(img)
if os.path.isfile(outfile2):
os.remove(outfile2)
hdu.writeto(outfile2)
def rotate_n_crop(ff_bin, ff_path, Flat_frame, Flat_frame_scalar):
""" Function for rotating the maxframe from bin file and cropping the meteor part based on the FTP_detectinfo detection data.
"""
if not ff_path[-1] == os.sep:
ff_path += os.sep
if not os.path.exists(ff_path):
print ff_path+" does not exist!"
return False
FTPdetect_file = ""
for line in os.listdir(ff_path):
if ("FTPdetectinfo_" in line) and (".txt" in line) and (not "_original" in line):
FTPdetect_file = line
break
ff_bin_path = ff_path+ff_bin
max_nomean_array = max_nomean(ff_bin_path, Flat_frame, Flat_frame_scalar)
saveImage(max_nomean_array, ff_bin_path+"_max_nomean.bmp", print_name = False)
max_bg_mean = int(np.mean(max_nomean_array))
print max_bg_mean
###MUST MAKE SOME SORT OF IMAGE MASKING HERE!!!!!!!!!!!!! Problem is when meteor is in the corner, then the lightcurve will be calculated with dark corners during rotation
coord_list, rot_angle = get_FTPdetect_coordinates(ff_path+FTPdetect_file, ff_bin)
nrows = len(max_nomean_array)
ncols = len(max_nomean_array[0])
crop_array = np.zeros(shape=(nrows, ncols), dtype=np.int) #Make a temporary 2D array which helps determine the crop coordinates after rotation
for coord in coord_list:
x = coord[0]
y = coord[1]
crop_array[x][y] = 255
crop_array = rotate(crop_array, -rot_angle+90, order = 0)
first_x, first_y, last_x, last_y = find_crop_size(crop_array)
saveImage(crop_array, 'test_croptest.bmp', print_name = False)
rotated_img = rotate(max_nomean_array, -rot_angle+90, order = 0)
rotated_img[rotated_img < 3] = max_bg_mean #Polish out the black edges
#print rotated_img
max_nomean_croped = rotated_img[first_y:last_y, first_x:last_x] #Crop out the image array
saveImage(max_nomean_array, 'test_raw.bmp', print_name = False)
saveImage(rotated_img, 'test_rotated.bmp', print_name = False)
saveImage(max_nomean_croped, 'test_meteor_croped.bmp', print_name = False)
return max_nomean_croped
def tiltaxisalign(im_series,tilt_angles,shift_and_tilt=('hold','hold')):
series_shape = np.shape(im_series)
new_series = im_series.copy()
final_series = im_series.copy()
#deg0_int = input('Which image is the 0 degree image? ')
midy = int(series_shape[1]/2)
axis_shift = shift_and_tilt[0]
axis_tilt = shift_and_tilt[1]
if axis_shift == 'hold':
shift_continue = 1
while shift_continue == 1:
plt.imshow(iradon(np.rot90(new_series[:,midy,:]), # rot
theta = tilt_angles,output_size = series_shape[2]))# anti-clokwise
plt.show()
axis_shift = float(input('By how many pixels from the original mid-point should the tilt axis be shifted? '))
for i in range(series_shape[0]):
new_series[i,:,:] = interpolation.shift(im_series.copy()[i,:,:],(0,axis_shift)) # shift along np x-axis
shift_continue = int(input('Would you like to apply further image shifts (1 for yes, 0 for no)? '))
for i in range(series_shape[0]):
final_series[i,:,:] = interpolation.shift(final_series[i,:,:],(0,axis_shift))
topy = int(3*series_shape[1]/8)
bottomy = int(5*series_shape[1]/8)
if axis_tilt == 'hold':
tilt_series = final_series.copy()
tilt_continue = 1
while tilt_continue == 1:
plt.imshow(iradon(np.rot90(new_series[:,topy,:]), theta = tilt_angles,output_size = series_shape[2]))
plt.show()
plt.imshow(iradon(np.rot90(new_series[:,bottomy,:]), theta = tilt_angles,output_size = series_shape[2]))
plt.show()
axis_tilt = float(input('By what angle from the original y axis (in degrees) should the tilt axis be rotated? '))
for i in range(series_shape[0]):
new_series[i,:,:] = interpolation.rotate(tilt_series.copy()[i,:,:],axis_tilt,reshape=False)
tilt_continue = int(input('Would you like to try another tilt angle (1 for yes, 0 for no)? '))
for i in range(series_shape[0]):
final_series[i,:,:] = interpolation.rotate(final_series[i,:,:],axis_tilt,reshape=False)
shift_and_tilt = (axis_shift,axis_tilt)
return(final_series, shift_and_tilt)
def scatter(self,move_pix=0,scale=1):
'''
Generate the scattered image which is stored in the ``iss`` member.
:param move_pix: (optional) int
Number of pixels to roll the screen (for time evolution).
:param scale: (optional) scalar
Scale factor for gradient. To simulate the scattering effect at another
wavelength this is (lambda_new/lambda_old)**2
'''
M = self.model.shape[-1] # size of original image array
N = self.nx # size of output image array
#if not self.live_dangerously: self._checkSanity()
# calculate phase gradient
dphi_x,dphi_y = self._calculate_dphi(move_pix=move_pix)
if scale != 1:
dphi_x *= scale/sqrt(2.)
dphi_y *= scale/sqrt(2.)
xx_,yy = np.meshgrid((np.arange(N) - 0.5*(N-1)),\
(np.arange(N) - 0.5*(N-1)),indexing='xy')
# check whether we care about PA of scattering kernel
if self.pa != None:
f_model = RectBivariateSpline(self.model_dx/self.dx*(np.arange(M) - 0.5*(M-1)),\
self.model_dx/self.dx*(np.arange(M) - 0.5*(M-1)),\
self.model)
# apply rotation
theta = -(90 * pi / 180) + np.radians(self.pa) # rotate CW 90 deg, then CCW by PA
xx_ += dphi_x
yy += dphi_y
xx = cos(theta)*xx_ - sin(theta)*yy
yy = sin(theta)*xx_ + cos(theta)*yy
self.iss = f_model.ev(yy.flatten(),xx.flatten()).reshape((self.nx,self.nx))
# rotate back and clip for positive values for I
if self.think_positive:
self.iss = clip(rotate(self.iss,-1*theta/np.pi*180,reshape=False),a_min=0,a_max=1e30) * (self.dx/self.model_dx)**2
else:
self.iss = rotate(self.iss,-1*theta/np.pi*180,reshape=False) * (self.dx/self.model_dx)**2
# otherwise do a faster lookup rather than the expensive interpolation.
else:
yyi = np.rint((yy+dphi_y+self.nx/2)).astype(np.int) % self.nx
xxi = np.rint((xx_+dphi_x+self.nx/2)).astype(np.int) % self.nx
if self.think_positive:
self.iss = clip(self.isrc[yyi,xxi],a_min=0,a_max=1e30)
else:
self.iss = self.isrc[yyi,xxi]
def computeOne(self, n):
for v in range(self._projections.views):
currentSm = self._systemMatrix.data[v]
angle = v*180/float(self._projections.views)
rotatedEstimate = interpolation.rotate(self._estimate, -angle, reshape=False)
backprojection = np.sum(currentSm * rotatedEstimate, axis=0)
normalization = np.sum(currentSm * currentSm, axis=0)
normalization[normalization == 0] = 1.0 #avoid division by zero
update = currentSm * (self._projections.data[v] - backprojection) / normalization
self._estimate += interpolation.rotate(update, angle, reshape=False)
self._result.append(self._estimate.copy())
def rotation_generator(generator, angle_range=(-180, 180)):
'''
yields rotated data and seg (rotated around center with a uniformly distributed angle between angle_range[0] and angle_range[1])
'''
for data, seg, labels in generator:
seg_min = np.min(seg)
seg_max = np.max(seg)
for sample_id in xrange(data.shape[0]):
angle = np.random.uniform(angle_range[0], angle_range[1])
data[sample_id] = interpolation.rotate(data[sample_id], angle, (1, 2), reshape=False, mode='nearest', order=3)
seg[sample_id] = np.round(interpolation.rotate(seg[sample_id], angle, (1, 2), reshape=False)).astype(np.int32)
seg[seg > seg_max] = seg_max
seg[seg < seg_min] = seg_min
yield data, seg, labels
def slitFoV(c, l, w, pa, instrument='vimos', os=20): if instrument=='vimos': frac = np.zeros((40,40)) elif instrument=='muse': frac = np.zeros((150,150)) # pa = np.radians(pa) s = frac.shape # sampler = np.array(np.meshgrid(np.arange(s[0]*100), np.arange(s[1]*100)) # ).T/100. sampler = np.zeros((s[0]*os,s[1]*os)) sampler[int(np.round(c[0] - w/2))*os: int(np.round(c[0] + w/2))*os, int(np.round(c[1] - l/2))*os: int(np.round(c[1] + l/2))*os] = 1 sampler = rotate(sampler, pa, reshape=False) i = np.arange(s[0]) j = np.arange(s[1]) i, j = np.meshgrid(i,j) i, j = i.flatten(), j.flatten() frac[j, i] = rebin(sampler, i*os+os/2, j*os+os/2, statistic='mean') return frac
def _augment_data(data):
image, rotations = data
augmented_images = [image] # Add the original image to the list
# Rotate and pad with 'nearest' pixels
augmented_images += [rotate(image.T, r, reshape=False, mode='nearest').T for r in rotations]
augmented_images = np.array(augmented_images)
return augmented_images
def Reconstruct(data, geom_params,size):
reconstructed = np.zeros((size,size))
for params in geom_params.values():
min_fs, max_fs, min_ss, max_ss = [int(p) for p in params[0:4]]
x = float(params[4].split('x')[0])
y = float(params[4].split('y')[0].split('x')[1])
data_tile = data[min_ss:max_ss+1,min_fs:max_fs+1][::-1,:]
delta_x = float(params[-2])
delta_y = float(params[-1])
if x > 0 and y > 0:
alpha = np.arctan(y/x) * 180. / np.pi
if x > 0 and y < 0:
alpha = np.arctan(y/x) * 180. / np.pi
if x < 0 and y > 0:
if np.abs(x) > 0.5:alpha = 180 - np.arctan(y/x) * 180. / np.pi
else: alpha = 180 + np.arctan(y/x) * 180. / np.pi
if x < 0 and y < 0:
alpha = 180 - np.arctan(y/np.abs(x)) * 180. / np.pi
rot = rotate(data_tile, alpha)
if int(alpha) in range(80,100):
reconstructed[(size / 2)-delta_y-rot.shape[0]:(size / 2)-delta_y,(size / 2)+delta_x-rot.shape[1]:(size / 2)+delta_x] = rot
elif int(alpha) in range(175,185):
reconstructed[(size / 2)-delta_y:(size / 2)-delta_y+rot.shape[0],(size / 2)+delta_x-rot.shape[1]:(size / 2)+delta_x] = rot
elif int(alpha) in range(-95,-85) or int(alpha) in range(265,275):
reconstructed[(size / 2)-delta_y:(size / 2)-delta_y+rot.shape[0],(size / 2)+delta_x:(size / 2)+delta_x+rot.shape[1]] = rot
else:
reconstructed[(size / 2)-delta_y-rot.shape[0]:(size / 2)-delta_y,(size / 2)+delta_x:(size / 2)+delta_x+rot.shape[1]] = rot
return reconstructed
def cellGridnessScore(rateMap, arenaDiam, h, corr_cutRmin):
'''
Compute a cell gridness score by taking the auto correlation of the
firing rate map, rotating it, and subtracting maxima of the
correlation coefficients of the former and latter, at 30, 90 and 150 (max),
and 60 and 120 deg. (minima). This gives the gridness score.
The center of the auto correlation map (given by corr_cutRmin) is removed
from the map
'''
rateMap_mean = rateMap - np.mean(np.reshape(rateMap, (1, rateMap.size)))
autoCorr, autoC_xedges, autoC_yedges = SNAutoCorr(rateMap_mean, arenaDiam, h)
# Remove the center point and
X, Y = np.meshgrid(autoC_xedges, autoC_yedges)
autoCorr[np.sqrt(X**2 + Y**2) < corr_cutRmin] = 0
da = 3
angles = list(range(0, 180+da, da))
crossCorr = []
# Rotate and compute correlation coefficient
for angle in angles:
autoCorrRot = rotate(autoCorr, angle, reshape=False)
C = np.corrcoef(np.reshape(autoCorr, (1, autoCorr.size)),
np.reshape(autoCorrRot, (1, autoCorrRot.size)))
crossCorr.append(C[0, 1])
max_angles_i = np.array([30, 90, 150]) / da
min_angles_i = np.array([60, 120]) / da
maxima = np.max(np.array(crossCorr)[max_angles_i])
minima = np.min(np.array(crossCorr)[min_angles_i])
G = minima - maxima
return G, np.array(crossCorr), angles
def drawRobot(self, mapObject, pos, val): robotMat = rotate(self.robotSprite, pos[2]+180) hgt = (robotMat.shape[0]-1)/2 # indices of center of robot wid = (robotMat.shape[1]-1)/2 x = slice(pos[0]-wid, pos[0]+wid+1, 1) # columns y = slice(pos[1]-hgt, pos[1]+hgt+1, 1) # rows mapObject[y,x][robotMat.astype(bool)] = val
def rotate(parangle,hdu):
"""Rotates fits file to match the rotation of the data cube.
Parameters
----------
parangle : float
Paralactic angle of the cube.
hdu : fits file
Returns
-------
hdu : fits file
Rotated fits file.
"""
mat = [[hdu[0].header['CD1_1'],hdu[0].header['CD1_2']],[hdu[0].header['CD2_1'],hdu[0].header['CD2_2']]]
CDELT1 = pow(abs(np.linalg.det(mat)),0.5) #Coordinate increment per pixel in DEGREES/PIXEL
sdssangle = np.arcsin(hdu[0].header['CD1_1']/CDELT1)*180./np.pi
hdu[0].data = interpolation.rotate(hdu[0].data,(parangle-sdssangle))
hdu[0].header['CD1_1']= np.sin(parangle*np.pi/180.)*CDELT1
hdu[0].header['CD1_2']= np.cos(parangle*np.pi/180.)*CDELT1
hdu[0].header['CD2_1']= np.cos(parangle*np.pi/180.)*CDELT1
hdu[0].header['CD2_2']= -np.sin(parangle*np.pi/180.)*CDELT1
hdu[0].header['CRPIX1']= hdu[0].data.shape[1]/2+0.5
hdu[0].header['CRPIX2']= hdu[0].data.shape[0]/2+0.5
hdu.writeto('imaging/LsdssDR12g.fits',clobber =True)
return hdu
def clip(ar, points, clipSize, if_Label):
'''
clips the input array to the given points. Because how the images are saved, a black border is created which can be removed by the function crop_around_center
'''
img = Image.new('F', (ar.shape[1], ar.shape[0]), 0)
#ImageDraw.Draw(img).polygon([X1, Y1, X2, Y2, X3, Y3, X4, Y4], outline=1, fill=1)
ImageDraw.Draw(img).polygon(points[0], outline=1, fill=1)
mask = np.array(img)
#create new image
newImArray = np.empty(ar.shape, dtype=np.float32)
#newImArray = ar * mask[:,:, np.newaxis]
if len(ar.shape) == 3:
newImArray = ar * mask[:,:,np.newaxis]
else:
newImArray = ar * mask[:,:]
croped_image = crop(newImArray, points[0], ar.shape)
newImArray = rotate(croped_image, points[1][0],\
reshape=False, order=0, mode='constant', cval=0.0)
if if_Label:
final = crop_around_center(newImArray, clipSize, points[1][0])
else:
final = crop_around_center(newImArray, clipSize, points[1][0])
#counter for fileIndex
#save_2_HDF5(inputFile, numberOfFiles, prefix, outputDir, channel, width, height, batch):
return (final)
def rotate(self,angle=0,transpose=False):
print 'Rotating data by %f degrees...' % angle
self.rotated = rotate(self.data,angle)
if transpose:
self.rotated = self.rotated.transpose()
return self.rotated
def apply_rotation(hdulist_or_filename=None, rotate_value=None, crop=True):
"""
Apply the detector's rotation to the PSF. This is for NIRCam, NIRISS, and FGS.
MIRI and NIRSpec's large rotation is already added inside WebbPSF's calculations.
Parameters
----------
hdulist_or_filename :
A PSF from WebbPSF, either as an HDUlist object or as a filename
rotate_value : float
Rotation in degrees that PSF needs to be. If set to None, function
will pull the most up to date SIAF value. Default = None.
crop : bool
True or False to crop the PSF so it matches the size of the input
PSF (e.g. so they could be more easily compared).
Returns
-------
psf : HDUlist object
PSF with rotation applied from SIAF values
"""
# Read in input PSF
if isinstance(hdulist_or_filename, str):
hdu_list = fits.open(hdulist_or_filename)
elif isinstance(hdulist_or_filename, fits.HDUList):
hdu_list = hdulist_or_filename
else:
raise ValueError("input must be a filename or HDUlist")
# Create a copy of the PSF
psf = copy.deepcopy(hdu_list)
# Log instrument and detector names
instrument = hdu_list[0].header["INSTRUME"].upper()
aper_name = hdu_list[0].header["APERNAME"].upper()
if instrument in ["MIRI", "NIRSPEC"]:
raise ValueError("{}'s rotation is already included in WebbPSF and "
"shouldn't be added again.".format(instrument))
# Set rotation value if not already set by a keyword argument
if rotate_value is None:
aper = _get_default_siaf(instrument, aper_name)
rotate_value = getattr(aper, "V3IdlYAngle") # the angle to rotate the PSF in degrees
# If crop = True, then reshape must be False - so invert this keyword
reshape = np.invert(crop)
ext = 1 # edit the oversampled PSF (OVERDIST extension)
psf_new = rotate(psf[ext].data, rotate_value, reshape=reshape)
# Apply data to correct extensions
psf[ext].data = psf_new
# Set new header keyword
psf[ext].header["ROTATION"] = (rotate_value, "PSF rotated to match detector rotation")
return psf
def getGeomTransformations(geom_params):
#reconstructed = np.zeros((size, size))
size = 1800
reconstructed = {}
min_fs, max_fs, min_ss, max_ss = [int(p) for p in geom_params[geom_params.keys()[0]][0:4]]
data = np.empty((max_ss - min_ss + 1, max_fs - min_fs + 1))
for key, params in geom_params.items():
#min_fs, max_fs, min_ss, max_ss = [int(p) for p in params[0:4]]
#print max_fs - min_fs + 1
#print max_ss - min_ss + 1
x = float(params[4].split('x')[0])
y = float(params[4].split('y')[0].split('x')[1])
#data_tile = data[ min_ss:max_ss + 1, min_fs:max_fs + 1][::-1,:] # , indices[1,::][min_ss:max_ss+1,min_fs:max_fs+1][::-1,:]
delta_x = float(params[-2])
delta_y = float(params[-1])
if x > 0 and y > 0:
alpha = np.arctan(y / x) * 180. / np.pi
if x > 0 and y < 0:
alpha = np.arctan(y / x) * 180. / np.pi
if x < 0 and y > 0:
if np.abs(x) > 0.5:
alpha = 180 - np.arctan(y / x) * 180. / np.pi
else:
alpha = 180 + np.arctan(y / x) * 180. / np.pi
if x < 0 and y < 0:
alpha = 180 - np.arctan(y / np.abs(x)) * 180. / np.pi
rot = rotate(data, alpha)
if int(alpha) in range(80, 100):
xmin = int(round((size / 2) - delta_y - rot.shape[0]))
xmax = int(round((size / 2) - delta_y))
ymin = int(round((size / 2) + delta_x - rot.shape[1]))
ymax = int(round((size / 2) + delta_x))
elif int(alpha) in range(175, 185):
xmin = int(round((size / 2) - delta_y))
xmax = int(round((size / 2) - delta_y + rot.shape[0]))
ymin = int(round((size / 2) + delta_x - rot.shape[1]))
ymax = int(round((size / 2) + delta_x))
elif int(alpha) in range(-95, -85) or int(alpha) in range(265, 275):
xmin = int(round((size / 2) - delta_y))
xmax = int(round((size / 2) - delta_y + rot.shape[0]))
ymin = int(round((size / 2) + delta_x))
ymax = int(round((size / 2) + delta_x + rot.shape[1]))
else:
xmin = int(round((size / 2) - delta_y - rot.shape[0]))
xmax = int(round((size / 2) - delta_y))
ymin = int(round((size / 2) + delta_x))
ymax = int(round((size / 2) + delta_x + rot.shape[1]))
reconstructed[key] = (alpha, xmin, xmax, ymin, ymax, delta_x, delta_y)
return reconstructed
def evaluate_rotate(rotations=[-5.0, -2.5, -1.0, 1, 2.5, 5.0], index=4, output_file='rotations.pdf'):
from scipy.ndimage import interpolation
import algorithms
original = face(index)
other = face(1)
faces = []
for rotation in rotations:
f = face(index)
f.abs_file.data['X'] = interpolation.rotate(f.abs_file.data['X'], rotation, mode='nearest', prefilter=False, reshape=False)
f.abs_file.data['Y'] = interpolation.rotate(f.abs_file.data['Y'], rotation, mode='nearest', prefilter=False, reshape=False)
f.abs_file.data['Z'] = interpolation.rotate(f.abs_file.data['Z'], rotation, mode='nearest', prefilter=False, reshape=False)
algorithms.features_histogram(f)
faces.append(f)
pyplot.figure()
subplot = pyplot.subplot(1, 2+len(rotations), 1)
subplot.imshow(original.Z)
subplot.title.set_text("Original")
subplot.title.set_fontsize(10)
subplot.xaxis.set_visible(False)
subplot.yaxis.set_visible(False)
for rotation, f, i in zip(rotations, faces, range(len(rotations))):
subplot = pyplot.subplot(1, 2+len(rotations), 2 + i)
subplot.imshow(f.Z)
subplot.title.set_text("%.1f deg" % rotation)
subplot.title.set_fontsize(10)
subplot.xaxis.set_visible(False)
subplot.yaxis.set_visible(False)
subplot = pyplot.subplot(1, 2+len(rotations), len(rotations) + 2)
subplot.imshow(other.Z)
subplot.title.set_text("Other")
subplot.title.set_fontsize(10)
subplot.xaxis.set_visible(False)
subplot.yaxis.set_visible(False)
pyplot.savefig(output_file, format='pdf', dpi=600, orientation='landscape', bbox_inches="tight")
return algorithms.similarity_matrix([original] + faces + [other], methods=[algorithms.distance_histogram_euclidean, algorithms.distance_histogram_city_block, algorithms.distance_histogram_correlation], normalizers=[False, False, False])
def estimate_skew_angle(image, angles):
estimates = []
for a in angles:
v = mean(interpolation.rotate(image, a, order=0, mode="constant"), axis=1)
v = var(v)
estimates.append((v, a))
_, a = max(estimates)
return a
def rotate_image(image, angle):
"""
:param image:
:param angle:
:return:
"""
return rotate(image, angle=angle, reshape=False)
def rotate_col(b_t, pre_theta, theta_t, n_theta, patch_size):
m, sparsity = b_t.shape
b_t_new = np.zeros_like(b_t)
for j in range(sparsity):
b_t_new[:, j] = rotate(b_t[:,j].reshape(patch_size), (theta_t[j]-pre_theta[j])*360./n_theta,
axes=(1, 0), reshape=False, order=3, mode='nearest').flatten()
return b_t_new
def plot_examples(X, y, y_angles, y_predicted_angles,
num_test_images=5, number=None, angle=None, only_errors=False):
if only_errors:
mask = np.where(y_predicted_angles != y_angles)[0]
X = X[mask]
y = y[mask]
y_predicted_angles = y_predicted_angles[mask]
y_angles = y_angles[mask]
if angle is not None and number is not None:
indices = np.intersect1d(np.where(y_angles == angle)[0], np.where(y == number)[0])
elif angle is not None:
indices = np.where(y_angles == angle)[0]
elif number is not None:
indices = np.where(y == number)[0]
else:
indices = len(y_angles)
mask = np.random.choice(indices, num_test_images)
true_angles = y_angles[mask]
predicted_angles = y_predicted_angles[mask]
plt.rcParams['figure.figsize'] = (10.0, 2 * num_test_images)
fig_number = 0
for i in range(num_test_images):
rotated_image = X[mask[i]][0]
original_image = rotate(rotated_image, -true_angles[i])
corrected_image = rotate(rotated_image, -predicted_angles[i])
fig_number += 1
plt.subplot(num_test_images, 3, fig_number)
plt.imshow(original_image)
fig_number += 1
plt.subplot(num_test_images, 3, fig_number)
plt.title('Angle: {0}'.format(true_angles[i]))
plt.imshow(rotated_image)
fig_number += 1
plt.subplot(num_test_images, 3, fig_number)
reconstructed_angle = angle_difference(predicted_angles[i], true_angles[i])
plt.title('Angle: {0}'.format(reconstructed_angle))
plt.imshow(corrected_image)
plt.tight_layout(pad=0.4, w_pad=0.5, h_pad=1.0)
def rotate90(): import matplotlib.pyplot as plt from scipy.ndimage.interpolation import rotate import scipy.misc image = io.imread(out_path + 'bibme0_contours.png') rotated_image = rotate(image, -90, reshape=True) toprint(rotated_image,"bibme0_contours.png")
def drawInset(self):
x, y = self.robot_pix[0:2] # indices of center of robot in main map
raw = int(self.insetSize_pix*0.75) # half the size of the main map segment to capture
rad = self.insetSize_pix/2 # half the size of the final segment
mapChunk = self.mapMatrix[y-raw:y+raw, x-raw:x+raw]
s = slice(raw-rad, raw+rad, 1) # region of rotated chunk that we want
self.insetMatrix = rotate(mapChunk, self.robot_rel[2], output=np.uint8, order=1, reshape=False)[s,s]
def detect_skew(img, min_angle=-20, max_angle=20, quality='low'):
img = sp.atleast_2d(img)
rows, cols = img.shape
min_min_angle = min_angle
max_max_angle = max_angle
if quality == 'low':
resolution = sp.arctan2(2.0, cols) * 180.0 / sp.pi
min_target_size = 100
resize_order = 1
elif quality == 'high':
resolution = sp.arctan2(1.0, cols) * 180.0 / sp.pi
min_target_size = 300
resize_order = 3
else:
resolution = sp.arctan2(1.0, cols) * 180.0 / sp.pi
min_target_size = 200
resize_order = 2
# resize the image so it's faster to work with
min_size = min(rows, cols)
target_size = min_target_size if min_size > min_target_size else min_size
resize_ratio = float(target_size) / min_size
img = imresize(img, resize_ratio)
rows, cols = img.shape
# pad the image and invert the colors
img *= -1
img += 255
padded_img = sp.zeros((rows*2, cols*2))
padded_img[rows//2:rows//2+rows, cols//2:cols//2+cols] = img
img = padded_img
# keep dividing the interval in half to achieve O(log(n))
while True:
current_resolution = (max_angle - min_angle) / 30.0
best_angle = None
best_variance = 0.0
# rotate the image, sum the pixel values in each row for each rotation
# then find the variance of all the sums, pick the highest variance
for i in xrange(31):
angle = min_angle + i * current_resolution
rotated_img = rotate(img, angle, reshape=False, order=resize_order)
num_black_pixels = sp.sum(rotated_img, axis=1)
variance = sp.var(num_black_pixels)
if variance > best_variance:
best_angle = angle
best_variance = variance
if current_resolution < resolution:
break
# update the angle range
min_angle = max(best_angle - current_resolution, min_min_angle)
max_angle = min(best_angle + current_resolution, max_max_angle)
return best_angle
def makeIMat(self, callback=None):
'''
Makes DM Interation Matrix
Initially, the DM influence functions are created using the method
``makeIMatShapes'', then if a rotation is specified these are rotated.
Each of the influence functions is passed to the specified ``WFS'' and
wfs measurements recorded.
Parameters:
callback (function): Function to be called on each WFS run
Returns:
ndarray: 2-dimensional interaction matrix
'''
logger.info("Making DM Influence Functions...")
self.makeIMatShapes()
# Imat value is in microns
# self.iMatShapes *= (self.dmConfig.iMatValue*1e-6)
if self.dmConfig.rotation:
self.iMatShapes = rotate(
self.iMatShapes, self.dmConfig.rotation,
order=self.dmConfig.interpOrder, axes=(-2,-1)
)
rotShape = self.iMatShapes.shape
self.iMatShapes = self.iMatShapes[:,
rotShape[1]/2. - self.simConfig.simSize/2.:
rotShape[1]/2. + self.simConfig.simSize/2.,
rotShape[2]/2. - self.simConfig.simSize/2.:
rotShape[2]/2. + self.simConfig.simSize/2.
]
iMat = numpy.zeros(
(self.iMatShapes.shape[0], self.totalWfsMeasurements) )
for i in xrange(self.iMatShapes.shape[0]):
subap=0
for nWfs in range(len(self.wfss)):
logger.debug("subap: {}".format(subap))
iMat[i, subap: subap + (2*self.wfss[nWfs].activeSubaps)] = (
self.wfss[nWfs].frame(
self.iMatShapes[i], iMatFrame=True
))
self.dmShape = self.iMatShapes[i]
if callback!=None:
callback()
logger.statusMessage(i, self.iMatShapes.shape[0],
"Generating {} Actuator DM iMat".format(self.acts))
subap += 2*self.wfss[nWfs].activeSubaps
self.iMat = iMat
return iMat
def __init__(self, filename, scale, flip_lr=False, rot_angle=None):
self.fn=filename[:8]
d=plt.imread(filename)
if flip_lr is True:
d=np.fliplr(d)
if rot_angle is not None:
d=rotate(d, rot_angle)
self.data=d
self.scale=scale
def __init__(self, flip_lr, rot_angle, multiply_by, scale):
if flip_lr is True:
self.d=np.fliplr(self.d)
if rot_angle is not None:
self.d=rotate(self.d, rot_angle)
self.rot_angle=rot_angle
self.data=self.d*multiply_by
self.scale=scale
self.s_name=os.path.basename(self.fn)[:8]
def find_scores(arr, angles):
scores = np.zeros_like(angles)
for i, a in enumerate(angles):
data = inter.rotate(arr, a, reshape=0, order=0)
hist = np.sum(data, axis=1)
scores[i] = np.sum((hist[1:] - hist[:-1]) ** 2)
return scores
def process(self):
for (n, input_file) in enumerate(self.input_files):
pcgts = page_from_file(self.workspace.download_file(input_file))
page_id = pcgts.pcGtsId or input_file.pageId or input_file.ID
page = pcgts.get_Page()
# why does it save the image ??
page_image, page_xywh, _ = self.workspace.image_from_page(
page, page_id)
if self.parameter['parallel'] < 2:
LOG.info("INPUT FILE %s ", input_file.pageId or input_file.ID)
raw = ocrolib.read_image_gray(page_image.filename)
flat = raw
#flat = np.array(binImg)
# estimate skew angle and rotate
if self.parameter['maxskew'] > 0:
if self.parameter['parallel'] < 2:
LOG.info("Estimating Skew Angle")
d0, d1 = flat.shape
o0, o1 = int(self.parameter['bignore'] * d0), int(
self.parameter['bignore'] * d1)
flat = amax(flat) - flat
flat -= amin(flat)
est = flat[o0:d0 - o0, o1:d1 - o1]
ma = self.parameter['maxskew']
ms = int(2 * self.parameter['maxskew'] *
self.parameter['skewsteps'])
angle = self.estimate_skew_angle(est,
linspace(-ma, ma, ms + 1))
flat = interpolation.rotate(flat,
angle,
mode='constant',
reshape=0)
flat = amax(flat) - flat
else:
angle = 0
# self.write_angles_to_pageXML(base,angle)
# estimate low and high thresholds
if self.parameter['parallel'] < 2:
LOG.info("Estimating Thresholds")
d0, d1 = flat.shape
o0, o1 = int(self.parameter['bignore'] * d0), int(
self.parameter['bignore'] * d1)
est = flat[o0:d0 - o0, o1:d1 - o1]
if self.parameter['escale'] > 0:
# by default, we use only regions that contain
# significant variance; this makes the percentile
# based low and high estimates more reliable
e = self.parameter['escale']
v = est - filters.gaussian_filter(est, e * 20.0)
v = filters.gaussian_filter(v**2, e * 20.0)**0.5
v = (v > 0.3 * amax(v))
v = morphology.binary_dilation(v,
structure=ones(
(int(e * 50), 1)))
v = morphology.binary_dilation(v,
structure=ones(
(1, int(e * 50))))
if self.parameter['debug'] > 0:
imshow(v)
ginput(1, self.parameter['debug'])
est = est[v]
lo = stats.scoreatpercentile(est.ravel(), self.parameter['lo'])
hi = stats.scoreatpercentile(est.ravel(), self.parameter['hi'])
# rescale the image to get the gray scale image
if self.parameter['parallel'] < 2:
LOG.info("Rescaling")
flat -= lo
flat /= (hi - lo)
flat = clip(flat, 0, 1)
if self.parameter['debug'] > 0:
imshow(flat, vmin=0, vmax=1)
ginput(1, self.parameter['debug'])
deskewed = 1 * (flat > self.parameter['threshold'])
# output the normalized grayscale and the thresholded images
LOG.info("%s lo-hi (%.2f %.2f) angle %4.1f" %
(pcgts.get_Page().imageFilename, lo, hi, angle))
if self.parameter['parallel'] < 2:
LOG.info("Writing")
#ocrolib.write_image_binary(base+".ds.png", deskewed)
#TODO: Need some clarification as the results effect the following pre-processing steps.
#orientation = -angle
#orientation = 180 - ((180 - orientation) % 360)
pcgts.get_Page().set_orientation(angle)
#print(orientation, angle)
file_id = input_file.ID.replace(self.input_file_grp,
self.output_file_grp)
if file_id == input_file.ID:
file_id = concat_padded(self.output_file_grp, n)
self.workspace.add_file(ID=file_id,
file_grp=self.output_file_grp,
pageId=input_file.pageId,
mimetype=MIMETYPE_PAGE,
local_filename=os.path.join(
self.output_file_grp,
file_id + '.xml'),
content=to_xml(pcgts).encode('utf-8'))
def rotate_image(image):
image=image.reshape((28, 28))
rotated_image=rotate(image, angle=180, cval=0, mode="constant")
return rotated_image.reshape((1, 28, 28, 1))
def level_leaf(array):
"""
takes the extrema of the leaf, finds the
hypotenuse, finds the angle between the base
and the hypotenuse, rotates the image by that
number, and reinitializes the leaf with the
new image
"""
edges_old = canny(array, 2.5)
thresh = threshold_otsu(array)
# binary = -(array > thresh)
binary = array <= thresh
binary = binary_erosion(
binary_closing(binary_dilation(binary_dilation(binary))))
edges = binary
left_extrema = [] # stem
right_extrema = [] # tip of leaf
# Create an array of index values
array_length = len(edges[0])
left_to_right = range(0, array_length)
for i in left_to_right:
column = edges[:, i]
# xvalue = np.argmax(self.edges[:, i])
if column.any():
left_extrema = [i, np.argmax(column)]
break
right_to_left = reversed(left_to_right)
for i in right_to_left:
column = edges[:, i]
if column.any():
right_extrema = [i, np.argmax(column)]
break
endpoints = [left_extrema, right_extrema]
left_endpoint, right_endpoint = endpoints
# find the distance (length) between the two points (hypotenuse)
diff_x = right_endpoint[0] - left_endpoint[0]
diff_y = right_endpoint[1] - left_endpoint[1]
hypot = sqrt((diff_x)**2 + (diff_y)**2)
# get the angle between the endpoints to rotate the image by
angle_radians = acos(diff_x / hypot)
angle_degrees = degrees(angle_radians)
array = array.copy()
# rotate the image, preserving size
if diff_y < 0:
array = rotate(array, -angle_degrees, reshape=True, mode='nearest')
else:
array = rotate(array, angle_degrees, reshape=True, mode='nearest')
# reinitzialize the image again
return array
def find_score(arr, angle):
data = inter.rotate(arr, angle, reshape=False, order=0)
hist = np.sum(data, axis=1)
score = np.sum((hist[1:] - hist[:-1])**2)
return hist, score
def yieldImages_ImgLabel(myobj):
# all the image are substrate by the mean and divided by its std for RGB channel, respectively.
#mask_process_f = get(myobj.ImageGenerator_Identifier)
#allDictList = getfileinfo(myobj.datadir, myobj.labelSuffix,myobj.dataExt,myobj.labelExt[0])
ImgList, ImgNameList = getfilelist(os.path.join(myobj.datadir, myobj.img),
myobj.dataExt)
#LabelList, LabelNameList = getfilelist(os.path.join(Imagefolder,'gt'), myobj.labelExt):
index_list = range(0, len(ImgList))
randshuffle(index_list)
for imgindx, thisindex in enumerate(index_list):
print('processing image {s}'.format(s=imgindx))
if imgindx == myobj.maximg:
break
thisimgfile = ImgList[thisindex]
thisimgname = ImgNameList[thisindex]
thismatfile = os.path.join(
myobj.datadir, myobj.gt,
thisimgname + myobj.labelSuffix[0] + myobj.labelExt[0])
#absmatfilelist.append(thismatfile)
#absfilelist.append(thisimgfile)
img_org = imread(thisimgfile) #np.asarray(Image.open(thisimgfile))
if os.path.isfile(thismatfile):
mask_org = imread(thismatfile) #loaded_mt = loadmat(thismatfile)
mask_org = (mask_org / np.max(mask_org))
else:
print(str(thismatfile) + ' does not exist!')
continue
filled_img = mask_org
for resizeratio in myobj.resizeratio:
#We may only interested in the region inside one region.
img_res = imresize(img_org, resizeratio)
mask_res = imresize(mask_org, resizeratio)
filled_img = imresize(mask_org, resizeratio)
#crop the boarder image
[rowsize, colsize] = [img_res.shape[0], img_res.shape[1]]
row_start = col_start = myobj.boarder * resizeratio
row_end = rowsize - myobj.boarder * resizeratio
col_end = colsize - myobj.boarder * resizeratio
img_res = img_res[row_start:row_end, col_start:col_end, ...]
mask_res = mask_res[row_start:row_end, col_start:col_end, ...]
filled_img = filled_img[row_start:row_end, col_start:col_end, ...]
for rotate_id in myobj.rotatepool:
if rotate_id == 0:
img_rot = img_res
mask_rot = mask_res
else:
img_rot = rotate(img_res, rotate_id, mode='nearest')
mask_rot = rotate(mask_res, rotate_id, mode='nearest')
mask = mask_rot
#assert np.max(mask.flatten()) <= 1, 'please normalize the mask to o, and 1 image'
img_rgb = img_rot.copy()
img = pre_process_img(img_rgb, yuv=False)
# if using special small patch, then crop
if myobj.crop_patch_size is None:
yield img, mask, filled_img
#break
else:
if myobj.crop_patch_size[0] <= 1:
crop_patch_size = (int(myobj.crop_patch_size[0] * mask.shape[0]), \
int(myobj.crop_patch_size[1] * mask.shape[1]))
else:
crop_patch_size = myobj.crop_patch_size
allcandidates = shuffle(find(mask != np.nan))
total_num = len(allcandidates)
selected_num = min(myobj.selected_num,
int(myobj.selected_portion * total_num))
selected_ind = allcandidates[0:selected_num]
Allpatches_vec_img = Points2Patches(
selected_ind, img, crop_patch_size)
Allpatches_vec_mask = Points2Patches(
selected_ind, mask, crop_patch_size)
Allpatches_vec_filled_img = Points2Patches(
selected_ind, filled_img, crop_patch_size)
AllPatches_img = np.reshape(
Allpatches_vec_img,
(selected_num, ) + crop_patch_size + (img.shape[2], ))
AllPatches_mask = np.reshape(
Allpatches_vec_mask,
(selected_num, ) + crop_patch_size + (mask.shape[2], ))
AllPatches_filled_img = np.reshape(
Allpatches_vec_filled_img,
(selected_num, ) + crop_patch_size + (mask.shape[2], ))
# imshow to check if the selected pts is correct.
for patch_idx in range(selected_num):
#imshow(AllPatches_img[patch_idx, ...])
yield AllPatches_img[patch_idx, ...], AllPatches_mask[
patch_idx, ...], AllPatches_filled_img[patch_idx,
...]
def random_rotation(image, angle_range=(0, 180)):
h, w, _ = image.shape
angle = np.random.randint(*angle_range)
image = rotate(image, angle)
image = resize(image, (h, w))
return image
def get_boxes(lows,data,size,lat,lon,landmask):
"""
box = get_boxes(lows, data, size)
Clips a square of length(2 x size) + 1 around each low
pressure center in lows and returns an array with all the
boxes.
Parameters:
--------------------
lows: binary matrix where 1 = low pressure center
data: numpy array, land masked with 0
size: numeric, half the length of the 2D subset box
edgedif: numeric, roughly the difference
in value between data and land grid cells
box: numpy array of data around low pressure centers
"""
lon[lon < 0.] = lon[lon < 0.] + 360.
long_size = ((size *2) + 1)
mylow = np.where(lows == 1)
nlows = mylow[0].shape[0]
data_box = np.zeros((nlows,long_size,long_size))
lat_box = np.zeros(data_box.shape)
lon_box = np.zeros(data_box.shape)
(tmax, ymax, xmax) = data.shape
if len(landmask.shape) == 3:
landmask = landmask[0,:,:]
# get lon where north is up
lon0 = lon[0,(int(xmax/2))-1]
count = 0
indlist = np.zeros((nlows))
for ind in range(0,nlows):
time = mylow[0][ind]
lowrow = mylow[1][ind]
lowcol = mylow[2][ind]
# -----------------
# rotation to north
# -----------------
mylon = lon[lowrow,lowcol]
low_mask = np.zeros((ymax,xmax))
low_mask[lowrow,lowcol] = 1
if lon0 < mylon:
deg = mylon - lon0
elif lon0 >= mylon:
deg = (360 + mylon) - lon0
low_rotated = interpolation.rotate(low_mask, deg, order = 2)
# because of interpolation, lows != 1
ynew,xnew = np.where(low_rotated == low_rotated.max())
if len(ynew.shape) > 1:
print("get_boxes: problem with rotation: too many indices for max")
print("get_boxes: exiting script")
#return
data_rotated = interpolation.rotate(data[time,:,:], deg, order =2)
# try to ignore data outside map
data_rotated[data_rotated == 0.0] = np.nan
# take out noisy grid cells near coast
landmask_rot = interpolation.rotate(landmask, deg, order = 2)
landmask_rot[landmask_rot < 0.5] = 0
landmask_rot[landmask_rot >= 0.5] = 1
data_rotated = buffer_coast(data_rotated, buf = 1, mask = landmask_rot)
# -----------------
# extracting box
# -----------------
y1 = int(ynew - size)
y2 = int(ynew + size + 1)
x1 = int(xnew - size)
x2 = int(xnew + size + 1)
if (y1 < 0) | (x1 < 0) | (y2 > ymax) | (x2 > xmax):
# too close to edge of map
continue
else:
data_box[count,:,:] = data_rotated[y1:y2,x1:x2]
#data_box[count,:,:] = data[ind,y1:y2,x1:x2]
#lat_box[count,:,:] = lat[y1:y2,x1:x2]
#lon_box[count,:,:] = lon[y1:y2,x1:x2]
indlist[count] = ind
count += 1
return data_box[0:count,:,:], indlist[0:count] #, lon_box[0:count,:,:]
def register(fn1, fn2, warpband, dims1=None, outfile=None):
gdal.AllRegister()
print '--------------------------------'
print' Register'
print'---------------------------------'
print time.asctime()
print 'reference image: '+fn1
print 'warp image: '+fn2
print 'warp band: %i'%warpband
start = time.time()
try:
if outfile is None:
path2 = os.path.dirname(fn2)
basename2 = os.path.basename(fn2)
root2, ext2 = os.path.splitext(basename2)
outfile = path2 + '/' + root2 + '_warp' + ext2
inDataset1 = gdal.Open(fn1,GA_ReadOnly)
inDataset2 = gdal.Open(fn2,GA_ReadOnly)
try:
cols1 = inDataset1.RasterXSize
rows1 = inDataset1.RasterYSize
cols2 = inDataset2.RasterXSize
rows2 = inDataset2.RasterYSize
bands2 = inDataset2.RasterCount
except Exception as e:
print 'Error %s --Image could not be read in'%e
sys.exit(1)
if dims1 is None:
x0 = 0
y0 = 0
else:
x0,y0,cols1,rows1 = dims1
band = inDataset1.GetRasterBand(warpband)
refband = band.ReadAsArray(x0,y0,cols1,rows1).astype(np.float32)
band = inDataset2.GetRasterBand(warpband)
warpband = band.ReadAsArray(x0,y0,cols1,rows1).astype(np.float32)
# similarity transform parameters for reference band number
scale, angle, shift = similarity(refband, warpband)
driver = inDataset2.GetDriver()
outDataset = driver.Create(outfile,cols1,rows1,bands2,GDT_Float32)
projection = inDataset1.GetProjection()
geotransform = inDataset1.GetGeoTransform()
if geotransform is not None:
gt = list(geotransform)
gt[0] = gt[0] + x0*gt[1]
gt[3] = gt[3] + y0*gt[5]
outDataset.SetGeoTransform(tuple(gt))
if projection is not None:
outDataset.SetProjection(projection)
# warp
for k in range(bands2):
inband = inDataset2.GetRasterBand(k+1)
outBand = outDataset.GetRasterBand(k+1)
bn1 = inband.ReadAsArray(0,0,cols2,rows2).astype(np.float32)
bn2 = ndii.zoom(bn1, 1.0 / scale)
bn2 = ndii.rotate(bn2, angle)
bn2 = ndii.shift(bn2, shift)
outBand.WriteArray(bn2[y0:y0+rows1, x0:x0+cols1])
outBand.FlushCache()
inDataset1 = None
inDataset2 = None
outDataset = None
print 'Warped image written to: %s'%outfile
print 'elapsed time: %s'%str(time.time()-start)
return outfile
except Exception as e:
print 'registersms failed: %s'%e
return None
flat = np.load(
'/scratch/dw1519/galex/data/star_photon/super_iterate{0}/flat.npy'.
format(i + 4))
profile = convert1(flat) / 0.6 * 0.75
mask2 = profile > 0
#profile[~mask2] = 1.
plt.plot(profile, '-', label='{0}-{1}'.format(bins[i], bins[i + 1]))
plt.legend()
#plt.ylim(0.5,1.)
plt.xlabel('x [pixel]')
plt.ylabel('Average sensitivity')
plt.savefig('/scratch/dw1519/galex/data/star_photon/profile_x.png',
dpi=190)
plt.clf()
flat0 = rotate(flat0, -90., reshape=False, order=1, prefilter=False)
profile0 = convert1(flat0)
mask = profile0 > 0
#profile0[~mask] = 1.
ratio = profile0 / profile0
plt.plot(profile0, '-', label='pipeline')
for i in range(6):
flat = np.load(
'/scratch/dw1519/galex/data/star_photon/super_iterate{0}/flat.npy'.
format(i + 4))
flat = rotate(flat, -90., reshape=False, order=1, prefilter=False)
profile = convert1(flat) / 0.6 * 0.72
mask2 = profile > 0
#profile[~mask2] = 1.
plt.plot(profile, '-', label='{0}-{1}'.format(bins[i], bins[i + 1]))
def decode_cv_prfs(n_pix, rsq_threshold, use_median, n_folds, data_file, extent, screen_distance, screen_width, TR, mask_name, **kwargs):
# for key, value in kwargs.iteritems():
# key = value
# set up results variables
cv_rotated_recon, cv_reshrot_recon, cv_reshrot_recon_m, \
cv_omega, cv_estimated_tau_matrix, \
cv_estimated_rho, cv_estimated_sigma, cv_estimated_alpha = [], [], [], [], [], [], [], []
for i in tqdm(range(n_folds)):
# get the data
(prf_cv_fold_data, W,
all_residuals_css, all_residual_covariance_css, test_data, mask) = setup_data_from_h5(
data_file = data_file,
n_pix=n_pix,
extent=extent,
screen_distance=screen_distance,
screen_width=screen_width,
rsq_threshold=rsq_threshold,
TR=TR,
cv_fold=i,
n_folds=n_folds,
use_median=False,
mask_name=mask_name)
# estimate the covariance structure, which outputs all parameters
(estimated_tau_matrix, estimated_rho,
estimated_sigma, estimated_alpha, omega, omega_inv, logdet) = fit_model_omega(observed_residual_covariance=all_residual_covariance_css,
WWT=np.dot(W,W.T),
verbose=0,
# infile='../data/omega.npy'
)
# set up result array:
# set up result array:
dm_pixel_logl_ratio = np.zeros((mask.sum(),test_data.shape[1]))
# and loop across timepoints
for t, bold in enumerate(test_data.T):
dm_pixel_logl_ratio[:,t] = firstpass_decoder_independent_channels(
W=W,
bold=bold,
logdet=logdet,
omega_inv=omega_inv,
mapping_relation=['power_law','linear'],
mapping_parameters=[prf_cv_fold_data[:, 3],prf_cv_fold_data[:,4:6]]
#mapping_relation='power_law',
#mapping_parameters=prf_cv_fold_data[:, 3]
)
decoded_image = np.zeros((mask.sum(),test_data.shape[1]))
for t, bold in enumerate(tqdm(test_data.T)):
logl, decoded_image[:,t] = maximize_loglikelihood( starting_value=dm_pixel_logl_ratio[:,t],
W=W,
bold=bold,
logdet=logdet,
omega_inv=omega_inv,
mapping_relation = ['power_law', 'linear'],
mapping_parameters = [prf_cv_fold_data[:, 3],prf_cv_fold_data[:,4:6]]
#mapping_relation='power_law',
#mapping_parameters=prf_cv_fold_data[:, 3]
)
# fill in the mask
recon = np.zeros([decoded_image.shape[1]]+list(mask.shape) )
for t in range(decoded_image.shape[1]):
recon[t,mask] = decoded_image[:,t]
# rotate reconstructions to bar orientation
thetas = [-1, 0, -1, 45, 270, -1, 315, 180, -1, 135, 90, -1, 225, -1]
rotated_recon = np.copy(recon).T
hrf_delay = 0
block_delimiters = np.r_[np.arange(2, 462, 34) + hrf_delay, 462]
reshrot_recon = np.zeros((8, rotated_recon.shape[0], rotated_recon.shape[1], 38))
bar_counter = 0
for i in range(len(block_delimiters) - 1):
if thetas[i] != -1:
rotated_recon[:, :, block_delimiters[i]:block_delimiters[i + 1] + 4] = rotate(rotated_recon[:, :, block_delimiters[i]:block_delimiters[i + 1] + 4],
axes=(
0, 1),
angle=thetas[i],
reshape=False,
mode='nearest')
reshrot_recon[bar_counter] = rotated_recon[:, :,
block_delimiters[i]:block_delimiters[i + 1] + 4]
bar_counter += 1
reshrot_recon_m = np.median(reshrot_recon, axis=0)
rotated_recon_m = np.median(rotated_recon, axis=0)
pl.figure(figsize=(24,7))
pl.imshow(rotated_recon_m);
pl.figure(figsize=(12,6))
pl.imshow(np.median(reshrot_recon_m, axis = 0), aspect = 0.5);
##############################
# Save out results
##############################
cv_rotated_recon.append(rotated_recon_m)
cv_reshrot_recon.append(reshrot_recon)
cv_reshrot_recon_m.append(reshrot_recon_m)
cv_omega.append(omega)
cv_estimated_tau_matrix.append(estimated_tau_matrix)
cv_estimated_rho.append(estimated_rho)
cv_estimated_sigma.append(estimated_sigma)
cv_estimated_alpha.append(estimated_alpha)
cv_rotated_recon = np.array(cv_rotated_recon)
cv_reshrot_recon = np.array(cv_reshrot_recon)
cv_reshrot_recon_m = np.array(cv_reshrot_recon_m)
cv_omega = np.array(cv_omega)
cv_estimated_tau_matrix = np.array(cv_estimated_tau_matrix)
cv_estimated_rho = np.array(cv_estimated_rho)
cv_estimated_sigma = np.array(cv_estimated_sigma)
cv_estimated_alpha = np.array(cv_estimated_alpha)
return cv_rotated_recon, cv_reshrot_recon, cv_reshrot_recon_m, cv_omega, cv_estimated_tau_matrix, cv_estimated_rho, cv_estimated_sigma, cv_estimated_alpha
# 0%| | 0/6 [00:00<?, ?it/s]
# data file found, returning local file ../data/V1.h5
# max tau: 140.139935703 min tau: 2.24584848951
# sigma: 5.75604779908 rho: 0.0140699975513
# summed squared distance: 444618382.496
# 17%|█▋ | 1/6 [02:01<10:05, 121.07s/it]
# data file found, returning local file ../data/V1.h5
# max tau: 43.8768334409 min tau: 0.830690536232
# sigma: 8.72147338743 rho: 0.111600068051
# summed squared distance: 3433452.07076
# 33%|███▎ | 2/6 [04:01<08:04, 121.02s/it]
# data file found, returning local file ../data/V1.h5
# max tau: 72.5664452049 min tau: 1.14343146174
# sigma: 5.92838085616 rho: 0.0351495921648
# summed squared distance: 35334222.2567
# 50%|█████ | 3/6 [05:55<05:56, 118.72s/it]
# data file found, returning local file ../data/V1.h5
# max tau: 41.4005198983 min tau: 0.790744766647
# sigma: -5.14436990217 rho: 0.125429150112
# summed squared distance: 2501341.73854
# 67%|██████▋ | 4/6 [07:47<03:53, 116.83s/it]
# data file found, returning local file ../data/V1.h5
# max tau: 39.3126611088 min tau: 2.16457366969
# sigma: 10.4218693161 rho: 0.0367255035114
# summed squared distance: 7811951.39835
# 83%|████████▎ | 5/6 [09:46<01:57, 117.32s/it]
# data file found, returning local file ../data/V1.h5
# max tau: 18.611556683 min tau: -0.360151297827
# sigma: 13.6759073136 rho: 0.232756200748
# summed squared distance: 209931.292302
# 100%|██████████| 6/6 [11:47<00:00, 118.61s/it]
short = img[~np.all(img == 255, axis=1)]
small = short.T[~np.all(short == 255, axis=0)].T
return small
img = clip_whitespace(img)
# To find a reasonable position for the code:
# 1. rotate a bit, trim whitespace, check width
# 2, minimum width is a reasonable rotation (nice_angle)
#NOTE: this is iterative search, consider binary or grqdient-like
#NOTE: this is exact to 1deg, consider making it more (or less) exact
size_sums = []
for angle in range(360):
rotated = interpolation.rotate(img, angle, cval=255)
clipped = clip_whitespace(rotated)
size_sums.append(clipped.shape[0])
nice_angle = np.array(size_sums).argmin()
rotated = interpolation.rotate(img, nice_angle, cval=255)
clipped = clip_whitespace(rotated)
qr = clipped
# Convert to a format known to the qr scanner and decode
# NOTE: the scanner is unfortunately too powerful and works for any orientation
# (in fact, it even works for the original image).
image = Image.fromarray(qr)
codes = zbarlight.scan_codes(['qrcode'], image)
code = str(codes[0], "utf-8")
print("Decoded: " + code)
def jacobi(a):
"""
Compute all eigenvalues and eigenvectors of a real symmetric matrix a of side-length n and output
d[0...n-1] that has the eigenvalues of a sorted into descending order while v[0...n-1][0...n-1] is a
matrix whose columns contain the corresponding nromalized eigenvectors. nrot is the number of the Jacobi rotations
that are required. Only the upper tirangle of a is accessed.
"""
eps = 1e-9 # (epsilon) accuracy
theta = np.radians(30)
nrot = 0
if np.size(a, 0) != np.size(a, 1):
print("Error: matrix must be symmertic. E.g., of shape nxn")
return
n = np.size(a, 0)
v = np.matrix([0] * n, [0] * n)
np.fill_diagonal(v, 1.0)
b = np.diagonal(
v) # b and d are diagonals that we initialize for convenience
d = np.diagonal(v)
z = np.zeros(n)
for i in range(1, 51):
sm = 0 # sum the magnitude of off-diagonal elements
for ip in range(0, n):
for iq in range(ip + 1, n + 1):
sm += abs(a[ip][iq])
if sm == 0:
eigsrt(d, v)
return v
if i < 4:
tresh = .2 * sm(n**2)
else:
tresh = 0
for ip in range(1, n):
for eq in range(ip + 1, n + 1):
g = 100 * abs(a[ip][iq])
if i > 4 and g <= eps * abs(d[ip]) and g <= eps * abs * d[
iq]: # after 4 sweeps, skip rotation if the off-diagonal element is small.
a[ip][iq] = 0
elif abs(a[ip][iq]) > tresh:
h = d[iq] - d[ip]
if g <= eps * abs(h):
t = a[ip][iq] / h
else:
theta = .5 * h / (a[ip][iq])
t = 1.0 / (abs(theta) + np.sqrt(1 + theta**2))
if theta < 0: # reverse direction
t = -t
c = 1.0 / np.sqrt(1 + t**2)
s = t * c
tau = s / (1.0 + c)
h = t * a[ip][iq]
z[ip] -= h
z[iq] += h
d[ip] -= h
a[ip][iq] = 0
# perform the rotations
for j in range(0, n + 1):
# not sure if I have to work case-by-case to handle the p and q parameters.
v = rotate(a, j)
for ip in range(0, n + 1):
b[ip] += z[ip]
d[ip] = b[ip]
z[ip] = 0
return v, nrot
def kepprf(infile,
plotfile,
rownum,
columns,
rows,
fluxes,
border,
background,
focus,
prfdir,
xtol,
ftol,
imscale,
colmap,
labcol,
apercol,
plt,
verbose,
logfile,
status,
cmdLine=False):
# input arguments
status = 0
seterr(all="ignore")
# log the call
hashline = '----------------------------------------------------------------------------'
kepmsg.log(logfile, hashline, verbose)
call = 'KEPPRF -- '
call += 'infile=' + infile + ' '
call += 'plotfile=' + plotfile + ' '
call += 'rownum=' + str(rownum) + ' '
call += 'columns=' + columns + ' '
call += 'rows=' + rows + ' '
call += 'fluxes=' + fluxes + ' '
call += 'border=' + str(border) + ' '
bground = 'n'
if (background): bground = 'y'
call += 'background=' + bground + ' '
focs = 'n'
if (focus): focs = 'y'
call += 'focus=' + focs + ' '
call += 'prfdir=' + prfdir + ' '
call += 'xtol=' + str(xtol) + ' '
call += 'ftol=' + str(xtol) + ' '
call += 'imscale=' + imscale + ' '
call += 'colmap=' + colmap + ' '
call += 'labcol=' + labcol + ' '
call += 'apercol=' + apercol + ' '
plotit = 'n'
if (plt): plotit = 'y'
call += 'plot=' + plotit + ' '
chatter = 'n'
if (verbose): chatter = 'y'
call += 'verbose=' + chatter + ' '
call += 'logfile=' + logfile
kepmsg.log(logfile, call + '\n', verbose)
# test log file
logfile = kepmsg.test(logfile)
# start time
kepmsg.clock('KEPPRF started at', logfile, verbose)
# reference color map
if colmap == 'browse':
status = cmap_plot(cmdLine)
# construct inital guess vector for fit
if status == 0:
guess = []
try:
f = fluxes.strip().split(',')
x = columns.strip().split(',')
y = rows.strip().split(',')
for i in range(len(f)):
f[i] = float(f[i])
except:
f = fluxes
x = columns
y = rows
nsrc = len(f)
for i in range(nsrc):
try:
guess.append(float(f[i]))
except:
message = 'ERROR -- KEPPRF: Fluxes must be floating point numbers'
status = kepmsg.err(logfile, message, verbose)
if status == 0:
if len(x) != nsrc or len(y) != nsrc:
message = 'ERROR -- KEPFIT:FITMULTIPRF: Guesses for rows, columns and '
message += 'fluxes must have the same number of sources'
status = kepmsg.err(logfile, message, verbose)
if status == 0:
for i in range(nsrc):
try:
guess.append(float(x[i]))
except:
message = 'ERROR -- KEPPRF: Columns must be floating point numbers'
status = kepmsg.err(logfile, message, verbose)
if status == 0:
for i in range(nsrc):
try:
guess.append(float(y[i]))
except:
message = 'ERROR -- KEPPRF: Rows must be floating point numbers'
status = kepmsg.err(logfile, message, verbose)
if status == 0 and background:
if border == 0:
guess.append(0.0)
else:
for i in range((border + 1) * 2):
guess.append(0.0)
if status == 0 and focus:
guess.append(1.0)
guess.append(1.0)
guess.append(0.0)
# open TPF FITS file
if status == 0:
try:
kepid, channel, skygroup, module, output, quarter, season, \
ra, dec, column, row, kepmag, xdim, ydim, barytime, status = \
kepio.readTPF(infile,'TIME',logfile,verbose)
except:
message = 'ERROR -- KEPPRF: is %s a Target Pixel File? ' % infile
status = kepmsg.err(logfile, message, verbose)
if status == 0:
kepid, channel, skygroup, module, output, quarter, season, \
ra, dec, column, row, kepmag, xdim, ydim, tcorr, status = \
kepio.readTPF(infile,'TIMECORR',logfile,verbose)
if status == 0:
kepid, channel, skygroup, module, output, quarter, season, \
ra, dec, column, row, kepmag, xdim, ydim, cadno, status = \
kepio.readTPF(infile,'CADENCENO',logfile,verbose)
if status == 0:
kepid, channel, skygroup, module, output, quarter, season, \
ra, dec, column, row, kepmag, xdim, ydim, fluxpixels, status = \
kepio.readTPF(infile,'FLUX',logfile,verbose)
if status == 0:
kepid, channel, skygroup, module, output, quarter, season, \
ra, dec, column, row, kepmag, xdim, ydim, errpixels, status = \
kepio.readTPF(infile,'FLUX_ERR',logfile,verbose)
if status == 0:
kepid, channel, skygroup, module, output, quarter, season, \
ra, dec, column, row, kepmag, xdim, ydim, qual, status = \
kepio.readTPF(infile,'QUALITY',logfile,verbose)
# read mask defintion data from TPF file
if status == 0:
maskimg, pixcoord1, pixcoord2, status = kepio.readMaskDefinition(
infile, logfile, verbose)
npix = numpy.size(numpy.nonzero(maskimg)[0])
# print target data
if status == 0 and verbose:
print('')
print(' KepID: %s' % kepid)
print(' BJD: %.2f' % (barytime[rownum - 1] + 2454833.0))
print(' RA (J2000): %s' % ra)
print('Dec (J2000): %s' % dec)
print(' KepMag: %s' % kepmag)
print(' SkyGroup: %2s' % skygroup)
print(' Season: %2s' % str(season))
print(' Channel: %2s' % channel)
print(' Module: %2s' % module)
print(' Output: %1s' % output)
print('')
# is this a good row with finite timestamp and pixels?
if status == 0:
if not numpy.isfinite(barytime[rownum - 1]) or numpy.nansum(
fluxpixels[rownum - 1, :]) == numpy.nan:
message = 'ERROR -- KEPFIELD: Row ' + str(
rownum) + ' is a bad quality timestamp'
status = kepmsg.err(logfile, message, verbose)
# construct input pixel image
if status == 0:
flux = fluxpixels[rownum - 1, :]
ferr = errpixels[rownum - 1, :]
DATx = arange(column, column + xdim)
DATy = arange(row, row + ydim)
# if numpy.nanmin > 420000.0: flux -= 420000.0
# image scale and intensity limits of pixel data
if status == 0:
n = 0
DATimg = empty((ydim, xdim))
ERRimg = empty((ydim, xdim))
for i in range(ydim):
for j in range(xdim):
DATimg[i, j] = flux[n]
ERRimg[i, j] = ferr[n]
n += 1
# determine suitable PRF calibration file
if status == 0:
if int(module) < 10:
prefix = 'kplr0'
else:
prefix = 'kplr'
prfglob = prfdir + '/' + prefix + str(module) + '.' + str(
output) + '*' + '_prf.fits'
try:
prffile = glob.glob(prfglob)[0]
except:
message = 'ERROR -- KEPPRF: No PRF file found in ' + prfdir
status = kepmsg.err(logfile, message, verbose)
# read PRF images
if status == 0:
prfn = [0, 0, 0, 0, 0]
crpix1p = numpy.zeros((5), dtype='float32')
crpix2p = numpy.zeros((5), dtype='float32')
crval1p = numpy.zeros((5), dtype='float32')
crval2p = numpy.zeros((5), dtype='float32')
cdelt1p = numpy.zeros((5), dtype='float32')
cdelt2p = numpy.zeros((5), dtype='float32')
for i in range(5):
prfn[i], crpix1p[i], crpix2p[i], crval1p[i], crval2p[i], cdelt1p[i], cdelt2p[i], status \
= kepio.readPRFimage(prffile,i+1,logfile,verbose)
prfn = array(prfn)
PRFx = arange(0.5, shape(prfn[0])[1] + 0.5)
PRFy = arange(0.5, shape(prfn[0])[0] + 0.5)
PRFx = (PRFx - size(PRFx) / 2) * cdelt1p[0]
PRFy = (PRFy - size(PRFy) / 2) * cdelt2p[0]
# interpolate the calibrated PRF shape to the target position
if status == 0:
prf = zeros(shape(prfn[0]), dtype='float32')
prfWeight = zeros((5), dtype='float32')
for i in range(5):
prfWeight[i] = sqrt((column - crval1p[i])**2 +
(row - crval2p[i])**2)
if prfWeight[i] == 0.0:
prfWeight[i] = 1.0e-6
prf = prf + prfn[i] / prfWeight[i]
prf = prf / nansum(prf) / cdelt1p[0] / cdelt2p[0]
# interpolate the calibrated PRF shape to the target position
# if status == 0:
# prf = zeros(shape(prfn[0,:,:]),dtype='float32')
# px = crval1p + len(PRFx) / 2 * cdelt1p[0]
# py = crval2p + len(PRFy) / 2 * cdelt2p[0]
# pp = [[px[0],py[0]],
# [px[1],py[1]],
# [px[2],py[2]],
# [px[3],py[3]],
# [px[4],py[4]]]
# for index,value in ndenumerate(prf):
# pz = prfn[:,index[0],index[1]]
# prf[index] = griddata(pp, pz, ([column], [row]), method='linear')
# print shape(prf)
# location of the data image centered on the PRF image (in PRF pixel units)
if status == 0:
prfDimY = int(ydim / cdelt1p[0])
prfDimX = int(xdim / cdelt2p[0])
PRFy0 = (shape(prf)[0] - prfDimY) / 2
PRFx0 = (shape(prf)[1] - prfDimX) / 2
# interpolation function over the PRF
if status == 0:
splineInterpolation = scipy.interpolate.RectBivariateSpline(
PRFx, PRFy, prf)
# construct mesh for background model
if status == 0 and background:
bx = numpy.arange(1., float(xdim + 1))
by = numpy.arange(1., float(ydim + 1))
xx, yy = numpy.meshgrid(numpy.linspace(bx.min(), bx.max(), xdim),
numpy.linspace(by.min(), by.max(), ydim))
# fit PRF model to pixel data
if status == 0:
start = time.time()
if focus and background:
args = (DATx, DATy, DATimg, ERRimg, nsrc, border, xx, yy,
splineInterpolation, float(x[0]), float(y[0]))
ans = fmin_powell(kepfunc.PRFwithFocusAndBackground,
guess,
args=args,
xtol=xtol,
ftol=ftol,
disp=False)
elif focus and not background:
args = (DATx, DATy, DATimg, ERRimg, nsrc, splineInterpolation,
float(x[0]), float(y[0]))
ans = fmin_powell(kepfunc.PRFwithFocus,
guess,
args=args,
xtol=xtol,
ftol=ftol,
disp=False)
elif background and not focus:
args = (DATx, DATy, DATimg, ERRimg, nsrc, border, xx, yy,
splineInterpolation, float(x[0]), float(y[0]))
ans = fmin_powell(kepfunc.PRFwithBackground,
guess,
args=args,
xtol=xtol,
ftol=ftol,
disp=False)
else:
args = (DATx, DATy, DATimg, ERRimg, nsrc, splineInterpolation,
float(x[0]), float(y[0]))
ans = fmin_powell(kepfunc.PRF,
guess,
args=args,
xtol=xtol,
ftol=ftol,
disp=False)
print('Convergence time = %.2fs\n' % (time.time() - start))
# pad the PRF data if the PRF array is smaller than the data array
if status == 0:
flux = []
OBJx = []
OBJy = []
PRFmod = numpy.zeros((prfDimY, prfDimX))
if PRFy0 < 0 or PRFx0 < 0.0:
PRFmod = numpy.zeros((prfDimY, prfDimX))
superPRF = zeros((prfDimY + 1, prfDimX + 1))
superPRF[abs(PRFy0):abs(PRFy0) + shape(prf)[0],
abs(PRFx0):abs(PRFx0) + shape(prf)[1]] = prf
prf = superPRF * 1.0
PRFy0 = 0
PRFx0 = 0
# rotate the PRF model around its center
if focus:
angle = ans[-1]
prf = rotate(prf, -angle, reshape=False, mode='nearest')
# iterate through the sources in the best fit PSF model
for i in range(nsrc):
flux.append(ans[i])
OBJx.append(ans[nsrc + i])
OBJy.append(ans[nsrc * 2 + i])
# calculate best-fit model
y = (OBJy[i] - mean(DATy)) / cdelt1p[0]
x = (OBJx[i] - mean(DATx)) / cdelt2p[0]
prfTmp = shift(prf, [y, x], order=3, mode='constant')
prfTmp = prfTmp[PRFy0:PRFy0 + prfDimY, PRFx0:PRFx0 + prfDimX]
PRFmod = PRFmod + prfTmp * flux[i]
wx = 1.0
wy = 1.0
angle = 0
b = 0.0
# write out best fit parameters
if verbose:
txt = 'Flux = %10.2f e-/s ' % flux[i]
txt += 'X = %9.4f pix ' % OBJx[i]
txt += 'Y = %9.4f pix ' % OBJy[i]
kepmsg.log(logfile, txt, True)
#
# params = {'backend': 'png',
# 'axes.linewidth': 2.5,
# 'axes.labelsize': 24,
# 'axes.font': 'sans-serif',
# 'axes.fontweight' : 'bold',
# 'text.fontsize': 12,
# 'legend.fontsize': 12,
# 'xtick.labelsize': 24,
# 'ytick.labelsize': 24}
# pylab.rcParams.update(params)
#
# pylab.figure(figsize=[20,10])
# ax = pylab.axes([0.05,0.08,0.46,0.9])
# xxx = numpy.arange(397.5,402.5,0.02)
# yyy = numpy.sum(PRFmod,axis=0) / numpy.max(numpy.sum(PRFmod,axis=0))
# pylab.plot(xxx,yyy,color='b',linewidth=3.0)
# xxx = numpy.append(numpy.insert(xxx,[0],[xxx[0]]),xxx[-1])
# yyy = numpy.append(numpy.insert(yyy,[0],[0.0]),yyy[-1])
# pylab.fill(xxx,yyy,fc='y',linewidth=0.0,alpha=0.3)
# pylab.xlabel('Pixel Column Number')
# pylab.xlim(397.5,402.5)
# pylab.ylim(1.0e-30,1.02)
# for xmaj in numpy.arange(397.5,402.5,1.0):
# pylab.plot([xmaj,xmaj],[0.0,1.1],color='k',linewidth=0.5,linestyle=':')
# for xmaj in numpy.arange(0.2,1.2,0.2):
# pylab.plot([0.0,2000.0],[xmaj,xmaj],color='k',linewidth=0.5,linestyle=':')
#
#
# ax = pylab.axes([0.51,0.08,0.46,0.9])
# xxx = numpy.arange(32.5,37.5,0.02)
# yyy = numpy.sum(PRFmod,axis=1) / numpy.max(numpy.sum(PRFmod,axis=1))
# pylab.plot(xxx,yyy,color='b',linewidth=3.0)
# xxx = numpy.append(numpy.insert(xxx,[0],[xxx[0]]),xxx[-1])
# yyy = numpy.append(numpy.insert(yyy,[0],[0.0]),yyy[-1])
# pylab.fill(xxx,yyy,fc='y',linewidth=0.0,alpha=0.3)
# pylab.setp(pylab.gca(),yticklabels=[])
# pylab.xlabel('Pixel Row Number')
# pylab.xlim(32.5,37.5)
# pylab.ylim(1.0e-30,1.02)
# for xmaj in numpy.arange(32.5,37.5,1.0):
# pylab.plot([xmaj,xmaj],[0.0,1.1],color='k',linewidth=0.5,linestyle=':')
# for xmaj in numpy.arange(0.2,1.2,0.2):
# pylab.plot([0.0,2000.0],[xmaj,xmaj],color='k',linewidth=0.5,linestyle=':')
# pylab.ion()
# pylab.plot([])
# pylab.ioff()
if verbose and background:
bterms = border + 1
if bterms == 1:
b = ans[nsrc * 3]
else:
bcoeff = array([
ans[nsrc * 3:nsrc * 3 + bterms],
ans[nsrc * 3 + bterms:nsrc * 3 + bterms * 2]
])
bkg = kepfunc.polyval2d(xx, yy, bcoeff)
b = nanmean(bkg.reshape(bkg.size))
txt = '\n Mean background = %.2f e-/s' % b
kepmsg.log(logfile, txt, True)
if focus:
wx = ans[-3]
wy = ans[-2]
angle = ans[-1]
if verbose and focus:
if not background: kepmsg.log(logfile, '', True)
kepmsg.log(logfile, ' X/Y focus factors = %.3f/%.3f' % (wx, wy),
True)
kepmsg.log(logfile, 'PRF rotation angle = %.2f deg' % angle, True)
# measure flux fraction and contamination
# LUGER: This looks horribly bugged. ``PRFall`` is certainly NOT the sum of the all the sources.
if status == 0:
PRFall = kepfunc.PRF2DET(flux, OBJx, OBJy, DATx, DATy, wx, wy, angle,
splineInterpolation)
PRFone = kepfunc.PRF2DET([flux[0]], [OBJx[0]], [OBJy[0]], DATx, DATy,
wx, wy, angle, splineInterpolation)
# LUGER: Add up contaminant fluxes
PRFcont = np.zeros_like(PRFone)
for ncont in range(1, len(flux)):
PRFcont += kepfunc.PRF2DET([flux[ncont]], [OBJx[ncont]],
[OBJy[ncont]], DATx, DATy, wx, wy,
angle, splineInterpolation)
PRFcont[np.where(PRFcont < 0)] = 0
FluxInMaskAll = numpy.nansum(PRFall)
FluxInMaskOne = numpy.nansum(PRFone)
FluxInAperAll = 0.0
FluxInAperOne = 0.0
FluxInAperAllTrue = 0.0
for i in range(1, ydim):
for j in range(1, xdim):
if kepstat.bitInBitmap(maskimg[i, j], 2):
FluxInAperAll += PRFall[i, j]
FluxInAperOne += PRFone[i, j]
FluxInAperAllTrue += PRFone[i, j] + PRFcont[i, j]
FluxFraction = FluxInAperOne / flux[0]
try:
Contamination = (FluxInAperAll - FluxInAperOne) / FluxInAperAll
except:
Contamination = 0.0
# LUGER: Pixel crowding metrics
Crowding = PRFone / (PRFone + PRFcont)
# LUGER: Optimal aperture crowding metric
CrowdAper = FluxInAperOne / FluxInAperAllTrue
kepmsg.log(
logfile,
'\n Total flux in mask = %.2f e-/s' % FluxInMaskAll,
True)
kepmsg.log(
logfile,
' Target flux in mask = %.2f e-/s' % FluxInMaskOne,
True)
kepmsg.log(
logfile,
' Total flux in aperture = %.2f e-/s' % FluxInAperAll,
True)
kepmsg.log(
logfile,
' Target flux in aperture = %.2f e-/s' % FluxInAperOne,
True)
kepmsg.log(
logfile, ' Target flux fraction in aperture = %.2f%%' %
(FluxFraction * 100.0), True)
kepmsg.log(
logfile, 'Contamination fraction in aperture = %.2f%%' %
(Contamination * 100.0), True)
kepmsg.log(logfile,
' Crowding metric in aperture = %.4f' % (CrowdAper),
True)
# constuct model PRF in detector coordinates
if status == 0:
PRFfit = PRFall + 0.0
if background and bterms == 1:
PRFfit = PRFall + b
if background and bterms > 1:
PRFfit = PRFall + bkg
# calculate residual of DATA - FIT
if status == 0:
PRFres = DATimg - PRFfit
FLUXres = numpy.nansum(PRFres) / npix
# calculate the sum squared difference between data and model
if status == 0:
Pearson = abs(numpy.nansum(numpy.square(DATimg - PRFfit) / PRFfit))
Chi2 = numpy.nansum(
numpy.square(DATimg - PRFfit) / numpy.square(ERRimg))
DegOfFreedom = npix - len(guess) - 1
try:
kepmsg.log(logfile, '\n Residual flux = %.2f e-/s' % FLUXres,
True)
kepmsg.log(
logfile, 'Pearson\'s chi^2 test = %d for %d dof' %
(Pearson, DegOfFreedom), True)
except:
pass
kepmsg.log(
logfile,
' Chi^2 test = %d for %d dof' % (Chi2, DegOfFreedom),
True)
# image scale and intensity limits for plotting images
if status == 0:
imgdat_pl, zminfl, zmaxfl = kepplot.intScale2D(DATimg, imscale)
imgprf_pl, zminpr, zmaxpr = kepplot.intScale2D(PRFmod, imscale)
imgfit_pl, zminfi, zmaxfi = kepplot.intScale2D(PRFfit, imscale)
imgres_pl, zminre, zmaxre = kepplot.intScale2D(PRFres, 'linear')
if imscale == 'linear':
zmaxpr *= 0.9
elif imscale == 'logarithmic':
zmaxpr = numpy.max(zmaxpr)
zminpr = zmaxpr / 2
# plot style
if status == 0:
pylab.figure(figsize=[12, 10])
pylab.clf()
plotimage(imgdat_pl, zminfl, zmaxfl, 1, row, column, xdim, ydim, 0.07,
0.53, 'observation', colmap, labcol)
# pylab.text(830.0,242.1,'A',horizontalalignment='center',verticalalignment='center',
# fontsize=28,fontweight=500,color='white')
# pylab.text(831.1,240.62,'B',horizontalalignment='center',verticalalignment='center',
# fontsize=28,fontweight=500,color='white')
# plotimage(imgprf_pl,0.0,zmaxpr/0.5,2,row,column,xdim,ydim,0.52,0.52,'model',colmap)
plotimage(imgprf_pl, zminpr, zmaxpr, 2, row, column, xdim, ydim, 0.44,
0.53, 'model', colmap, labcol)
kepplot.borders(maskimg, xdim, ydim, pixcoord1, pixcoord2, 1, apercol,
'--', 0.5)
kepplot.borders(maskimg, xdim, ydim, pixcoord1, pixcoord2, 2, apercol,
'-', 3.0)
plotimage(imgfit_pl,
zminfl,
zmaxfl,
3,
row,
column,
xdim,
ydim,
0.07,
0.08,
'fit',
colmap,
labcol,
crowd=Crowding)
# plotimage(imgres_pl,-zmaxre,zmaxre,4,row,column,xdim,ydim,0.44,0.08,'residual',colmap,'k')
plotimage(imgres_pl, zminfl, zmaxfl, 4, row, column, xdim, ydim, 0.44,
0.08, 'residual', colmap, labcol)
# plot data color bar
# barwin = pylab.axes([0.84,0.53,0.06,0.45])
barwin = pylab.axes([0.84, 0.08, 0.06, 0.9])
if imscale == 'linear':
brange = numpy.arange(zminfl, zmaxfl, (zmaxfl - zminfl) / 1000)
elif imscale == 'logarithmic':
brange = numpy.arange(10.0**zminfl, 10.0**zmaxfl,
(10.0**zmaxfl - 10.0**zminfl) / 1000)
elif imscale == 'squareroot':
brange = numpy.arange(zminfl**2, zmaxfl**2,
(zmaxfl**2 - zminfl**2) / 1000)
if imscale == 'linear':
barimg = numpy.resize(brange, (1000, 1))
elif imscale == 'logarithmic':
barimg = numpy.log10(numpy.resize(brange, (1000, 1)))
elif imscale == 'squareroot':
barimg = numpy.sqrt(numpy.resize(brange, (1000, 1)))
try:
nrm = len(str(int(numpy.nanmax(brange)))) - 1
except:
nrm = 0
brange = brange / 10**nrm
pylab.imshow(barimg,
aspect='auto',
interpolation='nearest',
origin='lower',
vmin=numpy.nanmin(barimg),
vmax=numpy.nanmax(barimg),
extent=(0.0, 1.0, brange[0], brange[-1]),
cmap=colmap)
barwin.yaxis.tick_right()
barwin.yaxis.set_label_position('right')
barwin.yaxis.set_major_locator(MaxNLocator(7))
pylab.gca().yaxis.set_major_formatter(
pylab.ScalarFormatter(useOffset=False))
pylab.gca().set_autoscale_on(False)
pylab.setp(pylab.gca(), xticklabels=[], xticks=[])
pylab.ylabel('Flux (10$^%d$ e$^-$ s$^{-1}$)' % nrm)
setp(barwin.get_yticklabels(), 'rotation', 90)
barwin.yaxis.set_major_formatter(ticker.FormatStrFormatter('%.1f'))
# plot residual color bar
# barwin = pylab.axes([0.84,0.08,0.06,0.45])
# Brange = numpy.arange(-zmaxre,zmaxre,(zmaxre+zmaxre)/1000)
# try:
# nrm = len(str(int(numpy.nanmax(brange))))-1
# except:
# nrm = 0
# brange = brange / 10**nrm
# barimg = numpy.resize(brange,(1000,1))
# pylab.imshow(barimg,aspect='auto',interpolation='nearest',origin='lower',
# vmin=brange[0],vmax=brange[-1],extent=(0.0,1.0,brange[0],brange[-1]),cmap=colmap)
# barwin.yaxis.tick_right()
# barwin.yaxis.set_label_position('right')
# barwin.yaxis.set_major_formatter(ticker.FormatStrFormatter('%.1f'))
# barwin.yaxis.set_major_locator(MaxNLocator(7))
# pylab.gca().yaxis.set_major_formatter(pylab.ScalarFormatter(useOffset=False))
# pylab.gca().set_autoscale_on(False)
# pylab.setp(pylab.gca(),xticklabels=[],xticks=[])
# pylab.ylabel('Residual (10$^%d$ e$^-$ s$^{-1}$)' % nrm)
# setp(barwin.get_yticklabels(), 'rotation', 90)
# render plot
if status == 0 and len(plotfile) > 0 and plotfile.lower() != 'none':
pylab.savefig(plotfile)
if status == 0 and plt:
if cmdLine:
pylab.show(block=True)
else:
pylab.ion()
pylab.plot([])
pylab.ioff()
# stop time
kepmsg.clock('\nKEPPRF ended at', logfile, verbose)
return
#plt.imshow(bin_img, cmap='gray')
#plt.savefig('binary.png')
delta = 1
limit = 5
angles = np.arange(-limit, limit + delta, delta)
scores = []
for angle in angles:
hist, score = find_score(bin_img, angle)
scores.append(score)
best_score = max(scores)
best_angle = angles[scores.index(best_score)]
print('Best angle for image' + str(i) + ': {}'.format(best_angle))
# correct skew
data = inter.rotate(bin_img, best_angle, reshape=False, order=0)
img_s = im.fromarray(255 - (255 * data).astype("uint8")).convert("RGB")
os.makedirs("../BDRP_Project/skew_corrected_images", exist_ok=True)
img_s.save(r'skew_corrected_images\\skew_corrected_img' + str(i) + '.jpg')
# img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
img_c = 'skew_corrected_images\\skew_corrected_img' + str(i) + '.jpg'
img1 = cv2.imread(img_c, cv2.IMREAD_COLOR)
img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
# rescaling image
img1 = cv2.resize(img1, None, fx=2, fy=2, interpolation=cv2.INTER_CUBIC)
#img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
# noise removal
img2 = cv2.GaussianBlur(img1, (5, 5), 0)
def correctSkews(angle1, angle2, array):
from scipy.ndimage.interpolation import rotate
rotated1 = rotate(array, angle1, mode='nearest', axes=(0, 1))
angle1rad = angle1 / 360 * 2 * np.pi
rotated2 = rotate(rotated1, angle2, mode='nearest', axes=(2, 0))
return rotated1, rotated2
def main():
gdal.AllRegister()
path = auxil.select_directory('Choose working directory')
if path:
os.chdir(path)
# MS image
file1 = auxil.select_infile(title='Choose MS image')
if file1:
inDataset1 = gdal.Open(file1, GA_ReadOnly)
cols = inDataset1.RasterXSize
rows = inDataset1.RasterYSize
bands = inDataset1.RasterCount
else:
return
pos1 = auxil.select_pos(bands)
if not pos1:
return
num_bands = len(pos1)
dims = auxil.select_dims([0, 0, cols, rows])
if dims:
x10, y10, cols1, rows1 = dims
else:
return
# PAN image
file2 = auxil.select_infile(title='Choose PAN image')
if file2:
inDataset2 = gdal.Open(file2, GA_ReadOnly)
bands = inDataset2.RasterCount
else:
return
if bands > 1:
print 'Must be a single band (panchromatic) image'
return
geotransform1 = inDataset1.GetGeoTransform()
geotransform2 = inDataset2.GetGeoTransform()
# outfile
outfile, fmt = auxil.select_outfilefmt()
if not outfile:
return
# resolution ratio
ratio = auxil.select_integer(4, 'Resolution ratio (2 or 4)')
if not ratio:
return
# MS registration band
k1 = auxil.select_integer(1, 'MS band for registration')
if not k1:
return
print '========================='
print ' ATWT Pansharpening'
print '========================='
print time.asctime()
print 'MS file: ' + file1
print 'PAN file: ' + file2
# read in MS image
band = inDataset1.GetRasterBand(1)
tmp = band.ReadAsArray(0, 0, 1, 1)
dt = tmp.dtype
MS = np.asarray(np.zeros((num_bands, rows1, cols1)), dtype=dt)
k = 0
for b in pos1:
band = inDataset1.GetRasterBand(b)
MS[k, :, :] = band.ReadAsArray(x10, y10, cols1, rows1)
k += 1
# if integer assume 11-bit quantization, otherwise must be byte
if MS.dtype == np.int16:
fact = 8.0
MS = auxil.byteStretch(MS, (0, 2**11))
else:
fact = 1.0
# read in corresponding spatial subset of PAN image
if (geotransform1 is None) or (geotransform2 is None):
print 'Image not georeferenced, aborting'
return
# upper left corner pixel in PAN
gt1 = list(geotransform1)
gt2 = list(geotransform2)
ulx1 = gt1[0] + x10 * gt1[1]
uly1 = gt1[3] + y10 * gt1[5]
x20 = int(round(((ulx1 - gt2[0]) / gt2[1])))
y20 = int(round(((uly1 - gt2[3]) / gt2[5])))
cols2 = cols1 * ratio
rows2 = rows1 * ratio
band = inDataset2.GetRasterBand(1)
PAN = band.ReadAsArray(x20, y20, cols2, rows2)
# if integer assume 11-bit quantization, otherwise must be byte
if PAN.dtype == np.int16:
PAN = auxil.byteStretch(PAN, (0, 2**11))
# out array
sharpened = np.zeros((num_bands, rows2, cols2), dtype=np.float32)
# compress PAN to resolution of MS image using DWT
panDWT = auxil.DWTArray(PAN, cols2, rows2)
r = ratio
while r > 1:
panDWT.filter()
r /= 2
bn0 = panDWT.get_quadrant(0)
# register (and subset) MS image to compressed PAN image using selected MSband
lines0, samples0 = bn0.shape
bn1 = MS[k1 - 1, :, :]
# register (and subset) MS image to compressed PAN image
(scale, angle, shift) = auxil.similarity(bn0, bn1)
tmp = np.zeros((num_bands, lines0, samples0))
for k in range(num_bands):
bn1 = MS[k, :, :]
bn2 = ndii.zoom(bn1, 1.0 / scale)
bn2 = ndii.rotate(bn2, angle)
bn2 = ndii.shift(bn2, shift)
tmp[k, :, :] = bn2[0:lines0, 0:samples0]
MS = tmp
smpl = np.random.randint(cols2 * rows2, size=100000)
print 'Wavelet correlations:'
# loop over MS bands
for k in range(num_bands):
msATWT = auxil.ATWTArray(PAN)
r = ratio
while r > 1:
msATWT.filter()
r /= 2
# sample PAN wavelet details
X = msATWT.get_band(msATWT.num_iter)
X = X.ravel()[smpl]
# resize the ms band to scale of the pan image
ms_band = ndii.zoom(MS[k, :, :], ratio)
# sample details of MS band
tmpATWT = auxil.ATWTArray(ms_band)
r = ratio
while r > 1:
tmpATWT.filter()
r /= 2
Y = tmpATWT.get_band(msATWT.num_iter)
Y = Y.ravel()[smpl]
# get band for injection
bnd = tmpATWT.get_band(0)
tmpATWT = None
aa, bb, R = auxil.orthoregress(X, Y)
print 'Band ' + str(k + 1) + ': %8.3f' % R
# inject the filtered MS band
msATWT.inject(bnd)
# normalize wavelet components and expand
msATWT.normalize(aa, bb)
r = ratio
while r > 1:
msATWT.invert()
r /= 2
sharpened[k, :, :] = msATWT.get_band(0)
sharpened *= fact # rescale dynamic range
msATWT = None
# write to disk
driver = gdal.GetDriverByName(fmt)
outDataset = driver.Create(outfile, cols2, rows2, num_bands, GDT_Float32)
gt1[0] += x10 * ratio
gt1[3] -= y10 * ratio
gt1[1] = gt2[1]
gt1[2] = gt2[2]
gt1[4] = gt2[4]
gt1[5] = gt2[5]
outDataset.SetGeoTransform(tuple(gt1))
projection1 = inDataset1.GetProjection()
if projection1 is not None:
outDataset.SetProjection(projection1)
for k in range(num_bands):
outBand = outDataset.GetRasterBand(k + 1)
outBand.WriteArray(sharpened[k, :, :], 0, 0)
outBand.FlushCache()
outDataset = None
print 'Result written to %s' % outfile
inDataset1 = None
inDataset2 = None
def return_score(arr, angle):
data = interpolation.rotate(arr, angle, reshape=False, order=0)
histogram = np.sum(data, axis=1)
score = np.sum((histogram[1:] - histogram[:-1])**2)
return histogram, score
def yieldImages(myobj):
# all the image are substrate by the mean and divided by its std for RGB channel, respectively.
mask_process_f = get(myobj.ImageGenerator_Identifier)
if hasattr(myobj, 'allDictList'):
allDictList = myobj.allDictList
else:
allDictList = getfileinfo(myobj.datadir, myobj.labelSuffix,
myobj.dataExt, myobj.labelExt[0])
index_list = range(0, len(allDictList))
rotationlist = myobj.resizeratio
randshuffle(index_list)
randshuffle(rotationlist)
for imgindx, thisindex in enumerate(index_list):
if imgindx == myobj.maximg:
break
returnDict = allDictList[thisindex]
thismatfile = returnDict['thismatfile']
thisimgfile = returnDict['thisfile']
print(thisimgfile)
img_org = np.asarray(Image.open(thisimgfile))
loaded_mt = loadmat(thismatfile)
if type(myobj.contourname) is not list:
myobj.contourname = [myobj.contourname]
contour_mat = None
for contourname in myobj.contourname:
if contourname in loaded_mt.keys():
contour_mat = loaded_mt[contourname].tolist()[0]
break
if not contour_mat:
contour_mat = loaded_mt.values()[0].tolist()[0]
print('check the mat keys, we use the first one default key: ' +
loaded_mt.keys()[0])
outputs_dict = dict()
for resizeratio in rotationlist:
img_res = imresize(img_org, resizeratio)
#print('start mask')
process_dict = mask_process_f(myobj,
contour_mat,
img_org.shape[0:2],
resizeratio=resizeratio)
#print('end mask')
mask_res = process_dict['mask']
filled_img_res = process_dict[
'filled_img'] if 'filled_img' in process_dict.keys(
) else mask_org
shed_mask_res = process_dict['shed']
#We may only interested in the region inside one region.
#since mask and filled_img are already resized version, only image need to be resized
#mask_res = mask_org #imresize(mask_org, resizeratio)
#crop the boarder image
[rowsize, colsize] = [img_res.shape[0], img_res.shape[1]]
if myobj.boarder < 1:
row_board = myobj.boarder * rowsize
col_board = myobj.boarder * colsize
elif len(myobj.boarder) == 2:
row_board, col_board = myobj.boarder
else:
row_board = col_board = myobj.boarder
row_start, col_start = row_board * resizeratio, col_board * resizeratio
row_end = rowsize - row_board * resizeratio
col_end = colsize - col_board * resizeratio
img_res = img_res[row_start:row_end, col_start:col_end, ...]
mask_res = mask_res[row_start:row_end, col_start:col_end, ...]
filled_img_res = filled_img_res[row_start:row_end,
col_start:col_end, ...]
shed_mask_res = shed_mask_res[row_start:row_end, col_start:col_end,
...]
if myobj.get_validation == True:
outputs_dict['img'] = pre_process_img(img_res.copy(),
yuv=False)
outputs_dict['mask'] = mask_res
outputs_dict['filled_img'] = filled_img_res
outputs_dict['shed_mask'] = shed_mask_res
yield outputs_dict
break
for rotate_id in myobj.rotatepool:
if rotate_id == 0:
img_rot = img_res.copy()
mask_rot = mask_res.copy()
shed_rot = shed_mask_res.copy()
filled_img_rot = filled_img_res.copy()
else:
img_rot = rotate(img_res.copy(),
rotate_id,
mode='reflect',
reshape=False)
mask_rot = rotate(mask_res.copy(),
rotate_id,
mode='reflect',
reshape=False)
shed_rot = rotate(shed_mask_res.copy(),
rotate_id,
mode='reflect',
reshape=False)
filled_img_rot = rotate(filled_img_res.copy(),
rotate_id,
mode='reflect',
reshape=False)
#assert np.max(mask.flatten()) <= 1, 'please normalize the mask to o, and 1 image'
img_rot = pre_process_img(img_rot, yuv=False)
# if using special small patch, then crop
if myobj.crop_patch_size is None:
outputs_dict['img'] = img_rot
outputs_dict['mask'] = mask_rot
outputs_dict['filled_img'] = filled_img_rot
outputs_dict['shed_mask'] = shed_rot
for item in first_flip_channel(myobj, outputs_dict):
yield item
#break
else:
if myobj.crop_patch_size[0] <= 1:
crop_patch_size = (int(myobj.crop_patch_size[0] * mask_rot.shape[0]), \
int(myobj.crop_patch_size[1] * mask_rot.shape[1]))
else:
crop_patch_size = myobj.crop_patch_size
allcandidates = shuffle(find(mask_rot != np.nan))
total_num = len(allcandidates)
selected_num = min(myobj.selected_num,
int(myobj.selected_portion * total_num))
selected_ind = allcandidates[0:selected_num]
Allpatches_vec_img = Points2Patches(
selected_ind, img_rot, crop_patch_size)
Allpatches_vec_mask = Points2Patches(
selected_ind, mask_rot, crop_patch_size)
Allpatches_vec_filled_img = Points2Patches(
selected_ind, filled_img_rot, crop_patch_size)
Allpatches_vec_shed_mask = Points2Patches(
selected_ind, shed_rot, crop_patch_size)
AllPatches_img = np.reshape(
Allpatches_vec_img, (selected_num, ) +
crop_patch_size + (img_rot.shape[2], ))
AllPatches_mask = np.reshape(
Allpatches_vec_mask, (selected_num, ) +
crop_patch_size + (mask_rot.shape[2], ))
AllPatches_filled_img = np.reshape(
Allpatches_vec_filled_img, (selected_num, ) +
crop_patch_size + (mask_rot.shape[2], ))
AllPatches_shed_mask = np.reshape(
Allpatches_vec_shed_mask, (selected_num, ) +
crop_patch_size + (mask_rot.shape[2], ))
# imshow to check if the selected pts is correct.
for patch_idx in range(selected_num):
#imshow(AllPatches_img[patch_idx, ...])
outputs_dict['img'] = AllPatches_img[patch_idx, ...]
outputs_dict['mask'] = AllPatches_mask[patch_idx, ...]
outputs_dict['filled_img'] = AllPatches_filled_img[
patch_idx, ...]
outputs_dict['shed_mask'] = AllPatches_shed_mask[
patch_idx, ...]
for item in first_flip_channel(myobj, outputs_dict):
yield item
def batch_function(self, image):
angle = self.get_random_variable('angle')
return rotate(image[0], angle, axes=self.axes, reshape=False), \
rotate(image[1], angle, axes=self.axes, order=0, cval=self.ml, reshape=False)
def RotateTopAxis(self, data, angle):
return rotate(data, angle, axes=(1, 2), reshape=False)
def similarity(im0, im1):
"""Return similarity transformed image im1 and transformation parameters.
Transformation parameters are: isotropic scale factor, rotation angle (in
degrees), and translation vector.
A similarity transformation is an affine transformation with isotropic
scale and without shear.
Limitations:
Image shapes must be equal and square.
All image areas must have same scale, rotation, and shift.
Scale change must be less than 1.8.
No subpixel precision.
"""
if im0.shape != im1.shape:
raise ValueError("Images must have same shapes.")
elif len(im0.shape) != 2:
raise ValueError("Images must be 2 dimensional.")
f0 = fftshift(abs(fft2(im0)))
f1 = fftshift(abs(fft2(im1)))
h = highpass(f0.shape)
f0 *= h
f1 *= h
del h
f0, log_base = logpolar(f0)
f1, log_base = logpolar(f1)
f0 = fft2(f0)
f1 = fft2(f1)
r0 = abs(f0) * abs(f1)
ir = abs(ifft2((f0 * f1.conjugate()) / r0))
i0, i1 = np.unravel_index(np.argmax(ir), ir.shape)
angle = 180.0 * i0 / ir.shape[0]
scale = log_base ** i1
if scale > 1.8:
ir = abs(ifft2((f1 * f0.conjugate()) / r0))
i0, i1 = np.unravel_index(np.argmax(ir), ir.shape)
angle = -180.0 * i0 / ir.shape[0]
scale = 1.0 / (log_base ** i1)
if scale > 1.8:
raise ValueError("Images are not compatible. Scale change > 1.8")
if angle < -90.0:
angle += 180.0
elif angle > 90.0:
angle -= 180.0
im2 = ndii.zoom(im1, 1.0/scale)
im2 = ndii.rotate(im2, angle)
if im2.shape < im0.shape:
t = np.zeros_like(im0)
t[:im2.shape[0], :im2.shape[1]] = im2
im2 = t
elif im2.shape > im0.shape:
im2 = im2[:im0.shape[0], :im0.shape[1]]
f0 = fft2(im0)
f1 = fft2(im2)
ir = abs(ifft2((f0 * f1.conjugate()) / (abs(f0) * abs(f1))))
t0, t1 = np.unravel_index(np.argmax(ir), ir.shape)
if t0 > f0.shape[0] // 2:
t0 -= f0.shape[0]
if t1 > f0.shape[1] // 2:
t1 -= f0.shape[1]
im2 = ndii.shift(im2, [t0, t1])
# correct parameters for ndimage's internal processing
if angle > 0.0:
d = int((int(im1.shape[1] / scale) * math.sin(math.radians(angle))))
t0, t1 = t1, d+t0
elif angle < 0.0:
d = int((int(im1.shape[0] / scale) * math.sin(math.radians(angle))))
t0, t1 = d+t1, d+t0
scale = (im1.shape[1] - 1) / (int(im1.shape[1] / scale) - 1)
return im2, scale, angle, [-t0, -t1]
RawData = sorted(glob.glob(folderMean + '20110410/*.fits'))
dateObs = (RawData[0])[-36:-21]
t1 = time.time()
# Rotating HMI continumm images
RawByH = RawData[0:80]
dcn1 = "ImaRot_"
print "Reading and rotating images..."
for i in RawByH:
print "Reading and rotating image -->", i[-36:-21]
data = (fits.open(i, checksum=True))[1].data
hdr = (fits.open(i, checksum=True))[1].header
rot_angle = hdr["crota2"]
im = rotate(data, rot_angle, mode="nearest", prefilter=False)
np.save(folderRot + dcn1 + i[-36:-21], im)
print 50 * "*"
print "End rotation section.. \n"
print "Time elapsed rotating images -->", time.time() - t1
print 50 * "*"
del (data)
del (hdr)
del (im)
# Charging rotated images to create submaps
RotData = sorted(glob.glob(folderRot + "*.npy"))
t2 = time.time()
def tefi(template,
search_image,
candidate_pixels,
best_scales,
best_angles,
scales=[0.5, 0.57, 0.66, 0.76, 0.87, 1.0],
upsampling=1,
thresh=100,
alpha=math.pi / 16,
use_percentile=True,
verbose=False):
"""
Template Matching Filter (Tefi) is the third and final filter for ciratefi.
For every candidate pixel, tefi rotates and scales the template image by the list
of scales and angles passed in (which, ideally are the output from cefi and rafi
respectively) and performs template match around the candidate pixels at the
approriate scale and rotation angle. Here, the scales, angles and candidate
points should be a parrallel array structure.
Any points with correlation strength over the threshold are returned as
the the strongest candidates for the image location. If knows the point
exists in one location, thresh should be 100 and use_percentile = True.
parameters
----------
template : ndarray
The input search template used to 'query' the destination
image
image : ndarray
The image or sub-image to be searched
candidate_pixels : ndarray
array of candidate pixels in tuples (y,x), best if
the pixel are the output of Cifi
best_scales : ndarray
The list of best fit scales for each candidate point, the length
should equal the length of the candidate point list
best_angles : ndarray
The list of best fit rotation for each candidate point in radians,
the length should equal the length of the candidate point list
upsampling : int
upsample degree
thresh : float
The correlation thresh hold above which a point will
be a first grade candidate point. If use_percentile=True
this will act as a percentile, for example, passing 90 means
keep values in the top 90th percentile
use_percentile : bool
If True (default), thresh is a percentile instead of a hard
strength value
alpha : float
A float between 0 & 2*pi, alpha list = np.arange(0, 2*pi, alpha)
verbose : bool
Set to True in order to output images and text describing the outputs. Can
cause a serious decrease in performance. False by default.
returns
-------
results : ndarray
array of pixel tuples (y,x) which over the threshold
"""
# check all inputs for validity, probably a better way to do this
if search_image.shape < template.shape:
raise ValueError(
'Template Image is smaller than Search Image for template of'
'size: {} and search image of size: {}'.format(
template.shape, search_image.shape))
candidate_pixels = np.asarray(candidate_pixels)
if not candidate_pixels.size or not np.any(candidate_pixels):
raise ValueError('cadidate pixel list is empty')
best_scales = np.asarray(best_scales, dtype=np.float32)
if not best_scales.size or not np.any(best_scales):
raise ValueError('best_scale list is empty')
if best_scales.shape != search_image.shape:
raise ValueError(
'Search image and scales must be of the same shape '
'got: best scales shape: {}, search image shape: {}'.format(
best_scales.shape, search_image.shape))
best_angles = np.asarray(best_angles, dtype=np.float32)
if not best_angles.size or not np.any(best_angles):
raise ValueError('Input best angle list is empty')
best_scales = np.asarray(best_scales, dtype=float)
if not best_scales.size or not np.any(best_scales):
raise ValueError('Input best_scales list is empty')
if thresh < -1. or thresh > 1. and not use_percentile:
raise ValueError(
'Thresholds must be in range [-1,1] when not using percentiles. Got: {}'
.format(thresh))
# Check inputs
if upsampling < 1:
raise ValueError('Upsampling must be >= 1, got {}'.format(upsampling))
tefi_coeffs = np.zeros(candidate_pixels.shape[0])
# if verbose, preallocate pixel data
if verbose: # pragma: no cover
image_pixels = np.zeros((search_image.shape[0], search_image.shape[1]))
# check for upsampling
if upsampling > 1:
u_template = zoom(template, upsampling, order=3)
u_search_image = zoom(search_image, upsampling, order=3)
alpha_list = np.arange(0, 2 * math.pi, alpha)
candidate_pixels *= int(upsampling)
# Tefi -- Template Matching Filter
for i in range(len(candidate_pixels)):
y, x = candidate_pixels[i]
try:
best_scale_idx = (np.where(
scales == best_scales[y // upsampling, x // upsampling]))[0][0]
best_alpha_idx = (np.where(
np.isclose(alpha_list, best_angles[i], atol=.01)))[0][0]
except:
tefi_coeffs[i] = 0
continue
tefi_scales = np.array(scales).take(range(best_scale_idx - 1,
best_scale_idx + 2),
mode='wrap')
tefi_alphas = alpha_list.take(range(best_alpha_idx - 1,
best_alpha_idx + 2),
mode='wrap')
scalesxalphas = util.cartesian([tefi_scales, tefi_alphas])
max_coeff = -math.inf
for j in range(scalesxalphas.shape[0]):
transformed_template = imresize(u_template, scalesxalphas[j][0])
transformed_template = rotate(transformed_template,
scalesxalphas[j][1])
y_window, x_window = (math.floor(
transformed_template.shape[0] /
2), math.floor(transformed_template.shape[1] / 2))
cropped_search = u_search_image[y - y_window:y + y_window + 1,
x - x_window:x + x_window + 1]
if (y < y_window or x < x_window
or cropped_search.shape < transformed_template.shape
or cropped_search.shape != transformed_template.shape):
score = -1
else:
result = cv2.matchTemplate(transformed_template.astype(
np.float32),
cropped_search.astype(np.float32),
method=cv2.TM_CCORR_NORMED)
score = np.average(result)
if score > max_coeff:
max_coeff = score
tefi_coeffs[i] = max_coeff
if verbose: # pragma: no cover
image_pixels[y // upsampling, x // upsampling] = max_coeff
if use_percentile:
thresh = np.percentile(tefi_coeffs, int(thresh))
result_points = candidate_pixels[np.where(tefi_coeffs >= thresh)]
result_coeffs = tefi_coeffs[np.where(tefi_coeffs >= thresh)]
x = result_points[0][1]
y = result_points[0][0]
ideal_y = u_search_image.shape[0] / 2
ideal_x = u_search_image.shape[1] / 2
if verbose: # pragma: no cover
plt.imshow(image_pixels, interpolation='none')
plt.scatter(y=y / upsampling, x=x / upsampling, c='w', s=80)
plt.show()
x = (ideal_x - x) / upsampling
y = (ideal_y - y) / upsampling
return x, y, result_coeffs[0]
def load_series(pn, crop_to, sort_by='rot', blur=0, verbose=False):
'''
loads all ptycho reconstructions in a folder and sorts by a parameter
Parameters
----------
pn : String pathname of folder
sort_by: 'rot' or 'step' parameter by which to sort the data
crop_to: Int size of cropped object reconstruction
blur: Int to pass as sigma to gaussian blur object phase. 0 = no blur
verbose: Bool to print detailed output
Returns
-------
d_s: hyperspy singnal2D object function
p_s: hyperspy singnal2D probe function
d_s_fft: hyperspy singnal2D fourier transfor of object function
rad_fft: hyperspy singnal1D object radial profile of d_s_fft
r_s: hyperspy signal1D object scan rotation
s_s: hyperspy signal1D object probe step size
e_s: hyperspy signal1D object final error value
Example usage
-------------
from epsic_tools.toolbox.ptychography.load_pycho_series import load_series
pn = r'Y:\2020\cm26481-1\processing\Merlin\20200130_80kV_graphene_600C_pty\cluster\processing\pycho'
d_s, p_s, d_s_fft, rad_fft, r_s, s_s, e_s = load_series(pn,sort_by = 'rot', crop_to = 80)
hs.plot.plot_signals([d_s,p_s,d_s_fft, rad_fft], navigator_list=[r_s,s_s, e_s,None])
to do
------
Break loading from hdf file into seperate functions
'''
pn = pn + '/*.hdf'
#build list of files
fp_list = glob.glob(pn)
len_dat = len(fp_list)
n = 0 # counter
# iterate through files
print(pn)
if verbose:
print(fp_list)
for this_fp in fp_list:
fj = this_fp[:-4] + '.json'
#open json file
with open(fj) as r:
params = json.load(r)
with h5py.File(this_fp, 'r') as d5:
#get phase data
dat = d5['entry_1']['process_1']['output_1']['object_phase']
dat = dat[0, 0, 0, 0, 0, :, :]
#get modulus data
dat_m = d5['entry_1']['process_1']['output_1']['object_modulus']
dat_m = dat_m[0, 0, 0, 0, 0, :, :]
#rotate
rot_angle = 90 - params['process']['common']['scan']['rotation']
dat = rotate(dat, rot_angle)
dat_m = rotate(dat_m, rot_angle)
#get probe
probe = np.array(
d5['entry_1']['process_1']['output_1']['probe_phase'])
#sum seperate coherent modes
probe = probe[:, :, 0, 0, 0, :, :].sum(axis=(0, 1))
probe_m = np.array(
d5['entry_1']['process_1']['output_1']['probe_modulus'])
probe_m = probe_m[:, :, 0, 0, 0, :, :].sum(axis=(0, 1))
#get complex probe
probe_c = np.array(d5['entry_1']['process_1']['output_1']['probe'])
probe_c = probe_c[:, :, 0, 0, 0, :, :].sum(axis=(0, 1))
#get error plot
error = np.array(d5['entry_1']['process_1']['output_1']['error'])
error = error[error != 0]
d5.close() # probably not necessary but just in case!
if n == 0:
#initiate arrays if first itteration - make bigger than fist data to account for change in size
shape_dat = int(10 * (np.ceil(dat.shape[0] / 10) + 1))
shape_probe = int(10 * (np.ceil(probe.shape[0] / 10) + 1))
if verbose == True:
print(n, len_dat, shape_dat, shape_dat)
dat_arr = np.empty(shape=(2, len_dat, shape_dat, shape_dat))
probe_arr = np.empty(shape=(2, len_dat, shape_probe, shape_probe))
rot_arr = np.empty(shape=len_dat)
step_arr = np.empty(shape=len_dat)
err_arr = np.empty(shape=len_dat)
#full_err_arr = []#np.empty( (len_dat, len(error),))
# calculate parameters to pad loaded data to same size as initiated array if necessary
dat_diff = shape_dat - dat.shape[0]
pad_dat = int(np.ceil(dat_diff / 2))
del_dat = int(pad_dat - np.floor(dat_diff / 2))
probe_diff = shape_probe - probe.shape[0]
pad_probe = int(np.ceil(probe_diff / 2))
del_probe = int(pad_probe - np.floor(probe_diff / 2))
# populate rotation, step and error array
rot_arr[n] = float(params['process']['common']['scan']['rotation'])
step_arr[n] = float(params['process']['common']['scan']['dR'][0])
err_arr[n] = error[-1]
#full_err_arr.append(error)
if verbose == True:
print(n, ' step : ', step_arr[n], ', rot : ', rot_arr[n],
', err : ', err_arr[n])
# load data into arrays (padded if necessary)
if pad_dat > 0:
dat_arr[0, n, :, :] = np.pad(dat[del_dat:, del_dat:], pad_dat,
'edge') #object phase
dat_arr[1, n, :, :] = np.pad(dat_m[del_dat:, del_dat:], pad_dat,
'edge') #object mod
else:
start_ind = int(-np.floor(dat_diff / 2))
end_ind = int(np.ceil(dat_diff / 2))
if end_ind == 0:
end_ind = dat.shape[0]
dat_arr[0, n, :, :] = dat[start_ind:end_ind,
start_ind:end_ind] #object phase
dat_arr[1, n, :, :] = dat_m[start_ind:end_ind,
start_ind:end_ind] #object mod
if pad_probe > 0:
probe_arr[0, n, :, :] = np.pad(probe[del_probe:, del_probe:],
pad_probe, 'edge') #probe phase
probe_arr[1, n, :, :] = np.pad(probe_m[del_probe:, del_probe:],
pad_probe, 'edge') #probe mod
else:
probe_start_ind = int(-np.floor(probe_diff / 2))
probe_end_ind = int(np.ceil(probe_diff / 2))
if probe_end_ind == 0:
probe_end_ind = probe.shape[0]
probe_arr[0, n, :, :] = probe[probe_start_ind:probe_end_ind,
probe_start_ind:
probe_end_ind] #probe phase
probe_arr[1, n, :, :] = probe_m[probe_start_ind:probe_end_ind,
probe_start_ind:
probe_end_ind] #probe mod
n = n + 1
# define structured array and sort
w_type = np.dtype([('rot', 'float'), ('step', 'float')])
w = np.empty(len(rot_arr), dtype=w_type)
w['rot'] = rot_arr
w['step'] = step_arr
if sort_by == 'rot':
sort_ind = np.argsort(w, order=('rot', 'step'))
elif sort_by == 'step':
sort_ind = np.argsort(w, order=('step', 'rot'))
print(sort_ind)
# re-order arrays
rot_sort = rot_arr[sort_ind]
step_sort = step_arr[sort_ind]
err_sort = err_arr[sort_ind]
dat_sort = dat_arr[:, sort_ind, :, :]
probe_sort = probe_arr[:, sort_ind, :, :]
#full_err_arr from list to padded array
#temp_arr = np.zeros([len(full_err_arr),len(max(full_err_arr,key = lambda x: len(x)))])
#for i,j in enumerate(full_err_arr):
# temp_arr[i][0:len(j)] = j
#full_err_arr = temp_arr
#full_err_sort = full_err_arr[sort_ind, :]
# unsorted to hs signals
'''
d = hs.signals.Signal2D(data = dat_arr)
p = hs.signals.Signal2D(data = probe_arr)
r = hs.signals.Signal1D(data = rot_arr)
s = hs.signals.Signal1D(data = step_arr)
e = hs.signals.Signal1D(data = err_arr)
'''
#sorted to hs signals
d_s = hs.signals.Signal2D(data=dat_sort)
p_s = hs.signals.Signal2D(data=probe_sort)
r_s = hs.signals.Signal1D(data=rot_sort)
s_s = hs.signals.Signal1D(data=step_sort)
e_s = hs.signals.Signal1D(data=err_sort)
#fe_s = hs.signals.Signal1D(data = full_err_sort)
#crop
'''
d.crop(axis = (2), start =int((shape_dat / 2) - (crop_to/2)), end = int((shape_dat / 2) + (crop_to/2) ))
d.crop(axis = (3), start =int((shape_dat / 2) - (crop_to/2)), end = int((shape_dat / 2) + (crop_to/2) ))
'''
d_s.crop(axis=(2),
start=int((shape_dat / 2) - (crop_to / 2)),
end=int((shape_dat / 2) + (crop_to / 2)))
d_s.crop(axis=(3),
start=int((shape_dat / 2) - (crop_to / 2)),
end=int((shape_dat / 2) + (crop_to / 2)))
#gaussian blur
print(d_s.data.shape)
print(type(d_s.data))
d_s.map(gaussian, sigma=blur)
# fft
'''
dat_fft = np.fft.fft2(d.data)
dat_fft = np.fft.fftshift(dat_fft)
d_fft = hs.signals.Signal2D(data = np.log10(np.abs(dat_fft)**2))
d_fft.data = np.flip(d_fft.data, axis = 0)
'''
dat_sort_fft = np.fft.fft2(d_s.data)
dat_sort_fft = np.fft.fftshift(dat_sort_fft)
d_s_fft = hs.signals.Signal2D(data=np.log10(np.abs(dat_sort_fft)**2))
d_s_fft.data = np.flip(d_s_fft.data, axis=0)
fft_mask = np.zeros_like(d_s_fft.data, dtype='bool')
fft_shape = fft_mask.shape
d_s_fft.inav[:, 0].data[:, int(fft_shape[-1] / 2), :] = 0
d_s_fft.inav[:, 0].data[:, :, int(fft_shape[-2] / 2)] = 0
rad_fft = radial_profile.radial_profile_stack(d_s_fft)
print(n, ' files loaded successfully')
return d_s, p_s, d_s_fft, rad_fft, r_s, s_s, e_s #, fe_s
def data_augentation(X, Y, data_gen_args, data_path_file_name):
if "horizontal_flip" in data_gen_args:
""" Flip image and groundtruth horizontally by a chance of 33%
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth
returns: two tensors, X, Y, which have the same dimensions as the input
"""
if data_gen_args["horizontal_flip"] == True & random.choice(
[True, False, False]) == True:
X = np.flip(X, len(np.shape(X)) - 1)
Y = np.flip(Y, len(np.shape(Y)) - 1)
if "vertical_flip" in data_gen_args:
""" Flip image and groundtruth vertically by a 33% chance
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth
returns: two tensors, X, Y, which have the same dimensions as the input
"""
if data_gen_args["vertical_flip"] == True & random.choice(
[True, False, False]) == True:
X = np.flip(X, len(np.shape(X)) - 2)
Y = np.flip(Y, len(np.shape(Y)) - 2)
if "rotation_range" in data_gen_args and data_gen_args[
"rotation_range"] > 0:
""" Rotate image and groundtruth by a random angle within the rotation range
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth
returns: two tensors, X, Y, which have the same dimensions as the input
"""
angle = np.random.choice(int(
data_gen_args["rotation_range"] * 100)) / 100
if len(np.shape(X)) - 1 == 3:
X = np.nan_to_num(
rotate(X, angle, mode='nearest', axes=(2, 3), reshape=False))
Y = np.nan_to_num(
rotate(Y, angle, mode='nearest', axes=(2, 3), reshape=False))
else:
X = np.nan_to_num(
rotate(X, angle, mode='nearest', axes=(1, 2), reshape=False))
Y = np.nan_to_num(
rotate(Y, angle, mode='nearest', axes=(1, 2), reshape=False))
# Zoom so that there are no empty edges
shape_X = np.shape(X)
shape_Y = np.shape(Y)
length_X = np.max(shape_X)
zoom_length_X = length_X / (np.cos(math.radians(angle)) +
np.sin(math.radians(angle)))
zoom_length_Y = length_X / (np.cos(math.radians(angle)) +
np.sin(math.radians(angle)))
X = X[...,
int((length_X - zoom_length_X) /
2):int(length_X - ((length_X - zoom_length_X) / 2)),
int((length_X - zoom_length_X) /
2):int(length_X - ((length_X - zoom_length_X) / 2)), :]
X = np.nan_to_num(sk.transform.resize(X, shape_X))
Y = Y[...,
int((length_X - zoom_length_Y) /
2):int(length_X - ((length_X - zoom_length_Y) / 2)),
int((length_X - zoom_length_Y) /
2):int(length_X - ((length_X - zoom_length_Y) / 2)), :]
Y = np.nan_to_num(sk.transform.resize(Y, shape_Y))
if "width_shift_range" in data_gen_args and data_gen_args[
"width_shift_range"] > 0 & random.choice([True, False]) == True:
""" Shift the image and groundtruth width by a number within the width shift range by a 50% chance
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth
width shift range: float, amound of width shift
returns: two tensors, X, Y, which have the same dimensions as the input
"""
width_shift = np.random.choice(
int(data_gen_args["width_shift_range"] * 100))
shape_X = np.shape(X)
shape_Y = np.shape(Y)
if len(np.shape(X)) - 1 == 3:
size = np.size(X[0, 0, :, 0, 0])
else:
size = np.size(X[0, 0, :, 0, ])
start_width_shift = np.random.choice(
int(size - (size - width_shift) + 1))
X = X[:, :,
start_width_shift:int(size - int(size * width_shift / 100)), ...]
X = np.nan_to_num(sk.transform.resize(X, shape_X))
Y = Y[:, :,
start_width_shift:int(size - int(size * width_shift / 100)), ...]
Y = np.nan_to_num(sk.transform.resize(Y, shape_Y))
if "height_shift_range" in data_gen_args and data_gen_args[
"height_shift_range"] > 0 & random.choice([True, False]) == True:
""" Shift the image and groundtruth height by a number within the width shift range by a 50% chance
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth
height shift range: float, amound of height shift
returns: two tensors, X, Y, which have the same dimensions as the input
"""
height_shift = np.random.choice(
int(data_gen_args["height_shift_range"] * 100))
shape_X = np.shape(X)
shape_Y = np.shape(Y)
if len(np.shape(X)) - 1 == 3:
size = np.size(X[0, 0, :, 0, 0])
else:
size = np.size(X[0, 0, :, 0, ])
start_heigth_shift = np.random.choice(
int(size - (size - height_shift) + 1))
X = X[...,
start_heigth_shift:int(size - int(size * height_shift / 100)), :]
X = np.nan_to_num(sk.transform.resize(X, shape_X))
Y = Y[...,
start_heigth_shift:int(size - int(size * height_shift / 100)), :]
Y = np.nan_to_num(sk.transform.resize(Y, shape_Y))
if "zoom_range" in data_gen_args and data_gen_args[
"zoom_range"] > 0 & random.choice([True, False]) == True:
""" Zooms the image and groundtruth to a random magnification in the zoom range and to a random position in the image by a 50% chance
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth
zoom range: float, amound of zoom
returns: two tensors, X, Y, which have the same dimensions as the input
"""
zoom = np.random.choice(int(data_gen_args["zoom_range"] * 100)) + 1
shape_X = np.shape(X)
shape_Y = np.shape(Y)
if len(np.shape(X)) - 1 == 3:
size = np.size(X[0, 0, :, 0, 0])
else:
size = np.size(X[0, 0, :, 0, ])
x_position = np.random.choice(zoom)
y_position = np.random.choice(zoom)
X = X[..., x_position:int(x_position + size - zoom),
y_position:int(y_position + size - zoom), :]
X = np.nan_to_num(sk.transform.resize(X, shape_X))
Y = Y[..., x_position:int(x_position + size - zoom),
y_position:int(y_position + size - zoom), :]
Y = np.nan_to_num(sk.transform.resize(Y, shape_Y))
if "gaussian_noise" in data_gen_args and data_gen_args[
"gaussian_noise"] > 0 and random.choice([True, False, False
]) == True:
""" Adds gaussian noise to the image by a one-third chance
# Arguments
X: tensor, the input image
gaussian_noise: float, amound of gaussian noise added
returns: one tensors, X, which have the same dimensions as the input
"""
value = data_gen_args["gaussian_noise"]
X = X + np.random.normal(0, value)
if "gaussian_blur_image" in data_gen_args and data_gen_args[
"gaussian_blur_image"] > 0 and random.choice([True, False, False
]) == True:
""" Blurs the input image in gaussian fashin by a 33% chance
# Arguments
X: tensor, the input image
gaussian_blur: float, amound of gaussian blur added
returns: one tensor, X, which have the same dimensions as the input
"""
value = data_gen_args["gaussian_blur_image"]
X = gaussian_filter(X, sigma=value)
if "gaussian_blur_label" in data_gen_args and data_gen_args[
"gaussian_blur_label"] > 0 and random.choice([True, False, False
]) == True:
""" Blurs the groundtruth image in gaussian fashin by a 33% chance
# Arguments
Y: tensor, the groundtruth image
gaussian_blur: float, amound of gaussian blur added
returns: one tensor, Y, which have the same dimensions as the input
"""
value = data_gen_args["gaussian_blur_label"]
Y = gaussian_filter(Y, sigma=value)
if "contrast_range" in data_gen_args and random.choice(
[True, False, False]) == True:
""" Increases or decreases the contrast of the input in by the range given by a 33% chance
# Arguments
X: tensor, the groundtruth image
contrast_range: float, amound of contrast added or removed
returns: one tensor, X, which have the same dimensions as the input
"""
range = np.random.uniform(-1, 1) * data_gen_args["contrast_range"] + 1
min_X = np.min(X)
X = (X - min_X) * range + min_X
if "brightness_range" in data_gen_args and random.choice([True, False
]) == True:
""" Increases or decreases the brightness of the input in by the range given by a 50% chance
# Arguments
X: tensor, the groundtruth image
brightness_range: float, amound of gaussian noise added
returns: one tensor, X, which have the same dimensions as the input
"""
range = np.random.uniform(-1, 1) * data_gen_args["brightness_range"]
X = X + range * X
if "threshold_background_image" in data_gen_args and random.choice(
[True, False, False]) == True:
""" Thresholds the input image at the mean and sets every pixelvalue below the mean to zero by a 33% chance
# Arguments
X: tensor, the groundtruth image
threshold_background_image: True or False
returns: one tensor, X, which have the same dimensions as the input
"""
mean_X = np.mean(X)
X[X < mean_X] = 0
if "threshold_background_groundtruth" in data_gen_args and random.choice(
[True, False, False]) == True:
""" Thresholds the groundtruth image at the mean and sets every pixelvalue below the mean to zero by a 33% chance
# Arguments
Y: tensor, the groundtruth image
threshold_background_groundtruth: True or False
returns: one tensor, Y, which have the same dimensions as the input
"""
mean_Y = np.mean(Y) * 0.8
Y[Y < mean_Y] = 0
if "binarize_mask" in data_gen_args and data_gen_args[
"binarize_mask"] == True:
""" Binarize the groundtruth image at the mean and sets every pixelvalue below the mean to zero and every above to one
# Arguments
Y: tensor, the groundtruth image
binarize_mask: True or False
returns: one tensor, Y, which have the same dimensions as the input
"""
mean_Y = np.mean(Y)
Y[Y < mean_Y] = 0
Y[Y >= mean_Y] = 1
if "save_augmented_images" in data_gen_args and data_gen_args[
"save_augmented_images"] == True:
""" save all augmented images to a folder named "Augmentations" in the project folder to make augmentations trackable
# Arguments
X: tensor, the input image
Y: tensor, the groundtruth image
save_augmented_images: True or False
returns: Saves the image pair as .png to the Augmentations folder
"""
data_path, file_name = os.path.split(data_path_file_name)
Aug_path = (data_path + '/Augmentations/')
os.makedirs(Aug_path, exist_ok=True)
#TODO: Check if the folder and saving really works with the path
title = file_name.split("'")[1]
title = os.path.splitext(title)[0]
#datacomb = np.concatenate((X, Y), axis=2).astype("uint8")
#sk.io.imsave(Aug_path + title, datacomb[0,...])
if len(np.shape(X)) == 5:
plot2images(X[0, 10, ..., 0], Y[0, 10, :, :, 0], Aug_path, title)
elif len(np.shape(X)) == 4 and np.shape(X)[-1] == 3:
plot2images(X[0, ...], Y[0, ...], Aug_path, title)
elif len(np.shape(X)) == 4 and np.shape(X)[-1] == 1:
plot2images(X[0, ..., 0], Y[0, :, :, 0], Aug_path, title)
#logging.info("Augmented Dimensions:", np.shape(X), np.shape(Y))
return X, Y
def main():
usage = '''
Usage:
-----------------------------------------------------------------------
python %s [-d spatialDimensions] [-p bandPositions [-r resolution ratio]
[-b registration band] msfilename panfilename
-----------------------------------------------------------------------
bandPositions and spatialDimensions are lists,
e.g., -p [1,2,3] -d [0,0,400,400]
Outfile name is msfilename_pan_dwt with same format as msfilename
Note: PAN image must completely overlap MS image subset chosen
-----------------------------------------------------''' %sys.argv[0]
options, args = getopt.getopt(sys.argv[1:],'hd:p:r:b:')
ratio = 4
dims1 = None
pos1 = None
k1 = 0
for option, value in options:
if option == '-h':
print usage
return
elif option == '-r':
ratio = eval(value)
elif option == '-d':
dims1 = eval(value)
elif option == '-p':
pos1 = eval(value)
elif option == '-b':
k1 = eval(value)-1
if len(args) != 2:
print 'Incorrect number of arguments'
print usage
sys.exit(1)
gdal.AllRegister()
file1 = args[0]
file2 = args[1]
path = os.path.dirname(file1)
basename1 = os.path.basename(file1)
root1, ext1 = os.path.splitext(basename1)
outfile = '%s/%s_pan_dwt%s'%(path,root1,ext1)
# MS image
inDataset1 = gdal.Open(file1,GA_ReadOnly)
try:
cols = inDataset1.RasterXSize
rows = inDataset1.RasterYSize
bands = inDataset1.RasterCount
except Exception as e:
print 'Error: %e --Image could not be read'%e
sys.exit(1)
if pos1 is None:
pos1 = range(1,bands+1)
num_bands = len(pos1)
if dims1 is None:
dims1 = [0,0,cols,rows]
x10,y10,cols1,rows1 = dims1
# PAN image
inDataset2 = gdal.Open(file2,GA_ReadOnly)
try:
bands = inDataset2.RasterCount
except Exception as e:
print 'Error: %e --Image could not be read'%e
sys.exit(1)
if bands>1:
print 'PAN image must be a single band'
sys.exit(1)
geotransform1 = inDataset1.GetGeoTransform()
geotransform2 = inDataset2.GetGeoTransform()
if (geotransform1 is None) or (geotransform2 is None):
print 'Image not georeferenced, aborting'
sys.exit(1)
print '========================='
print ' DWT Pansharpening'
print '========================='
print time.asctime()
print 'MS file: '+file1
print 'PAN file: '+file2
# image arrays
band = inDataset1.GetRasterBand(1)
tmp = band.ReadAsArray(0,0,1,1)
dt = tmp.dtype
MS = np.asarray(np.zeros((num_bands,rows1,cols1)),dtype=dt)
k = 0
for b in pos1:
band = inDataset1.GetRasterBand(b)
MS[k,:,:] = band.ReadAsArray(x10,y10,cols1,rows1)
k += 1
# if integer assume 11bit quantization otherwise must be byte
if MS.dtype == np.int16:
fact = 8.0
MS = auxil.byteStretch(MS,(0,2**11))
else:
fact = 1.0
# read in corresponding spatial subset of PAN image
if (geotransform1 is None) or (geotransform2 is None):
print 'Image not georeferenced, aborting'
return
# upper left corner pixel in PAN
gt1 = list(geotransform1)
gt2 = list(geotransform2)
ulx1 = gt1[0] + x10*gt1[1]
uly1 = gt1[3] + y10*gt1[5]
x20 = int(round(((ulx1 - gt2[0])/gt2[1])))
y20 = int(round(((uly1 - gt2[3])/gt2[5])))
cols2 = cols1*ratio
rows2 = rows1*ratio
band = inDataset2.GetRasterBand(1)
PAN = band.ReadAsArray(x20,y20,cols2,rows2)
# if integer assume 11-bit quantization, otherwise must be byte
if PAN.dtype == np.int16:
PAN = auxil.byteStretch(PAN,(0,2**11))
# compress PAN to resolution of MS image
panDWT = auxil.DWTArray(PAN,cols2,rows2)
r = ratio
while r > 1:
panDWT.filter()
r /= 2
bn0 = panDWT.get_quadrant(0)
lines0,samples0 = bn0.shape
bn1 = MS[k1,:,:]
# register (and subset) MS image to compressed PAN image
(scale,angle,shift) = auxil.similarity(bn0,bn1)
tmp = np.zeros((num_bands,lines0,samples0))
for k in range(num_bands):
bn1 = MS[k,:,:]
bn2 = ndii.zoom(bn1, 1.0/scale)
bn2 = ndii.rotate(bn2, angle)
bn2 = ndii.shift(bn2, shift)
tmp[k,:,:] = bn2[0:lines0,0:samples0]
MS = tmp
# compress pan once more, extract wavelet quadrants, and restore
panDWT.filter()
fgpan = panDWT.get_quadrant(1)
gfpan = panDWT.get_quadrant(2)
ggpan = panDWT.get_quadrant(3)
panDWT.invert()
# output array
sharpened = np.zeros((num_bands,rows2,cols2),dtype=np.float32)
aa = np.zeros(3)
bb = np.zeros(3)
print 'Wavelet correlations:'
for i in range(num_bands):
# make copy of panDWT and inject ith ms band
msDWT = copy.deepcopy(panDWT)
msDWT.put_quadrant(MS[i,:,:],0)
# compress once more
msDWT.filter()
# determine wavelet normalization coefficents
ms = msDWT.get_quadrant(1)
aa[0],bb[0],R = auxil.orthoregress(fgpan.ravel(), ms.ravel())
Rs = 'Band '+str(i+1)+': %8.3f'%R
ms = msDWT.get_quadrant(2)
aa[1],bb[1],R = auxil.orthoregress(gfpan.ravel(), ms.ravel())
Rs += '%8.3f'%R
ms = msDWT.get_quadrant(3)
aa[2],bb[2],R = auxil.orthoregress(ggpan.ravel(), ms.ravel())
Rs += '%8.3f'%R
print Rs
# restore once and normalize wavelet coefficients
msDWT.invert()
msDWT.normalize(aa,bb)
# restore completely and collect result
r = 1
while r < ratio:
msDWT.invert()
r *= 2
sharpened[i,:,:] = msDWT.get_quadrant(0)
sharpened *= fact
# write to disk
driver = inDataset1.GetDriver()
outDataset = driver.Create(outfile,cols2,rows2,num_bands,GDT_Float32)
projection1 = inDataset1.GetProjection()
if projection1 is not None:
outDataset.SetProjection(projection1)
gt1 = list(geotransform1)
gt1[0] += x10*ratio
gt1[3] -= y10*ratio
gt1[1] = gt2[1]
gt1[2] = gt2[2]
gt1[4] = gt2[4]
gt1[5] = gt2[5]
outDataset.SetGeoTransform(tuple(gt1))
for k in range(num_bands):
outBand = outDataset.GetRasterBand(k+1)
outBand.WriteArray(sharpened[k,:,:],0,0)
outBand.FlushCache()
outDataset = None
print 'Result written to %s'%outfile
inDataset1 = None
inDataset2 = None
def cnn_training(X_train, y_train, X_val, y_val,
fc_units=[500,200,100], conv_featmap=[6,16,32],
l2_norm=0.01,
seed=235,
learning_rate=1e-2,
epoch=20,
batch_size=245,
verbose=False,
pre_trained_model=None,imglen=164):
print("Building Network Parameters: ")
print("fc_units={}".format(fc_units))
print("l2_norm={}".format(l2_norm))
print("seed={}".format(seed))
print("learning_rate={}".format(learning_rate))
# define the variables and parameter needed during training
with tf.name_scope('inputs'):
xs = tf.placeholder(shape=[None,imglen,imglen,4], dtype=tf.float32)
ys = tf.placeholder(shape=[None, ], dtype=tf.int64)
#xs,hmm=augment(xs,ys,horizontal_flip=True,rotate=3,crop_probability=0.4)
output,loss=CNN(xs,ys,img_len=imglen, channel_num=4, output_size=2,conv_featmap=conv_featmap,fc_units=fc_units,conv_kernel_size=[3,3], pooling_size=[2,2],l2_norm=l2_norm, seed=seed)
iters = int(X_train.shape[0] / batch_size)
print('number of batches for training: {}'.format(iters))
step = train_step(loss)
eve,pred = evaluate(output, ys)
iter_total = 0
best_acc = 0
cur_model_name = 'cnn_{}'.format(int(time.time()))
with tf.Session() as sess:
merge = tf.summary.merge_all()
writer = tf.summary.FileWriter("log/{}".format(cur_model_name), sess.graph)
saver = tf.train.Saver()
sess.run(tf.global_variables_initializer())
# try to restore the pre_trained
if pre_trained_model is not None:
try:
print("Load the model from: {}".format(pre_trained_model))
saver.restore(sess, 'model/{}'.format(pre_trained_model))
except Exception:
print("Load model Failed!")
pass
store_trainacc=[]
store_valacc=[]
for epc in range(epoch):
print("epoch {} ".format(epc + 1))
for itr in range(iters):
iter_total += 1
rng=randint(1,100)
training_batch_x = X_train[itr * batch_size: (1 + itr) * batch_size]
training_batch_y = y_train[itr * batch_size: (1 + itr) * batch_size]
if rng>60 & rng<=75:
training_batch_x=translate(training_batch_x,2,2)
elif rng>75 & rng<=90:
training_batch_x=rotate(training_batch_x,angle=3)
elif rng>90 & rng<=95:
training_batch_x=flip(training_batch_x,mode='h')
elif rng>95 & rng<=100:
training_batch_x=flip(training_batch_x,mode='v')
else:
pass
_, cur_loss = sess.run([step, loss], feed_dict={xs: training_batch_x, ys: training_batch_y})
if itr == iters-1:
# do validation
train_eve=sess.run(eve,feed_dict={xs:X_train,ys:y_train})
train_acc=100-train_eve*100/y_train.shape[0]
store_trainacc.append(train_acc)
valid_eve, merge_result = sess.run([eve, merge], feed_dict={xs: X_val, ys: y_val})
valid_acc = 100 - valid_eve * 100 / y_val.shape[0]
store_valacc.append(valid_acc)
#y=graph.get_tensor_by_name("evaluate/ArgMax:0")
result=sess.run(pred,feed_dict={xs:X_val})
percentage=sess.run(output,feed_dict={xs:X_val,ys: y_val})
if verbose:
print('loss: {} validation accuracy : {}%'.format(
cur_loss,
valid_acc))
# save the merge result summary
writer.add_summary(merge_result, iter_total)
# when achieve the best validation accuracy, we store the model paramters
if valid_acc > best_acc:
print('Best validation accuracy! iteration:{} accuracy: {}%'.format(iter_total, valid_acc))
best_acc = valid_acc
saver.save(sess, 'model/{}'.format(cur_model_name))
store_pred=result
store_truelabel=y_val
print("Traning ends. The best valid accuracy is {}. Model named {}.".format(best_acc, cur_model_name))
return best_acc,store_trainacc,store_valacc,store_pred,store_truelabel,percentage
def gridness(rate_map, box_xlen, box_ylen, return_acorr=False,
step_size=0.1*pq.m):
'''Calculates gridness of a rate map. Calculates the normalized
autocorrelation (A) of a rate map B where A is given as
A = 1/n\Sum_{x,y}(B - \bar{B})^{2}/\sigma_{B}^{2}. Further, the Pearsson's
product-moment correlation coefficients is calculated between A and A_{rot}
rotated 30 and 60 degrees. Finally the gridness is calculated as the
difference between the minimum of coefficients at 60 degrees and the
maximum of coefficients at 30 degrees i.e. gridness = min(r60) - max(r30).
In order to focus the analysis on symmetry of A the the central and the
outer part of the gridness is maximized by increasingly mask A at steps of
``step_size``. This function is inspired by Lukas Solankas gridcells
package from Matt Nolans lab.
Parameters
----------
rate_map : numpy.ndarray
box_xlen : quantities scalar in m
side length of quadratic box
step_size : quantities scalar in m
step size in masking
return_acorr : bool
return autocorrelation map or not
Returns
-------
out : gridness, (autocorrelation map)
'''
from scipy.ndimage.interpolation import rotate
import numpy.ma as ma
from exana.misc.tools import (is_quantities, fftcorrelate2d,
masked_corrcoef2d)
is_quantities([box_xlen, box_ylen, step_size], 'scalar')
box_xlen = box_xlen.rescale('m').magnitude
box_ylen = box_ylen.rescale('m').magnitude
step_size = step_size.rescale('m').magnitude
tmp_map = rate_map.copy()
tmp_map[~np.isfinite(tmp_map)] = 0
acorr = fftcorrelate2d(tmp_map, tmp_map, mode='full', normalize=True)
rows, cols = acorr.shape
b_x = np.linspace(-box_xlen/2., box_xlen/2., rows)
b_y = np.linspace(-box_ylen/2., box_ylen/2., cols)
B_x, B_y = np.meshgrid(b_x, b_y)
grids = []
acorrs = []
# TODO find size of middle gaussian and exclude
for outer in np.arange(box_xlen/4, box_xlen/2, step_size):
m_acorr = ma.masked_array(acorr, mask=np.sqrt(B_x**2 + B_y**2) > outer)
for inner in np.arange(0, box_xlen/4, step_size):
m_acorr = \
ma.masked_array(m_acorr, mask=np.sqrt(B_x**2 + B_y**2) < inner)
angles = range(30, 180+30, 30)
corr = []
# Rotate and compute correlation coefficient
for angle in angles:
rot_acorr = rotate(m_acorr, angle, reshape=False)
corr.append(masked_corrcoef2d(rot_acorr, m_acorr)[0, 1])
r60 = corr[1::2]
r30 = corr[::2]
grids.append(np.min(r60) - np.max(r30))
acorrs.append(m_acorr)
if return_acorr:
return max(grids), acorr, # acorrs[grids.index(max(grids))]
else:
return max(grids)
def rotateImage(self, angle):
self.image = rotate(self.image, angle=angle, reshape=False, cval=-1000)
self.dicomRotation = angle
def rotate2d(img, angle, axes=(0,1), reshape=False, interpolation=1, border_mode='constant', value=0):
return sci.rotate(img, angle, axes, reshape=reshape, order=interpolation, mode=border_mode, cval=value)
