def imstretch(self):
data = np.clip(self.data_array, self.threshold[0], self.threshold[1])
if self.mode == "linear":
pass
elif self.mode == "logarithmic":
data = np.reciprocal(1 + np.power(0.5 / data, self.factor))
elif self.mode == "gamma":
data = np.power(data, self.factor)
elif self.mode == "arcsinh":
mn = np.nanmin(data)
mx = np.nanmax(data)
tmp = bytescale(data, high=1.0)
beta = np.clip(self.factor, 0.0, self.factor)
sclbeta = (beta - mn) / (mx - mn)
sclbeta = np.clip(sclbeta, 1.0e-12, sclbeta)
nonlinearity = 1.0 / sclbeta
extrema = np.arcsinh(np.array([0.0, nonlinearity]))
data = np.clip(np.arcsinh(data * nonlinearity), extrema[0], extrema[1])
elif self.mode == "square root":
data = np.sqrt(np.fabs(data)) * np.sign(data)
elif self.mode == "histogram equalization":
imhist, bins = np.histogram(data.flatten(), 256, normed=True)
cdf = imhist.cumsum() # cumulative distribution function
cdf = 255 * cdf / cdf[-1] # normalize
im2 = np.interp(data.flatten(), bins[:-1], cdf)
data = im2.reshape(data.shape)
self.scaled = bytescale(data).flatten().tolist()
def test_bytescale_keywords(self):
x = np.array([40, 60, 120, 200, 300, 500])
res_lowhigh = misc.bytescale(x, low=10, high=143)
assert_equal(res_lowhigh, [10, 16, 33, 56, 85, 143])
res_cmincmax = misc.bytescale(x, cmin=60, cmax=300)
assert_equal(res_cmincmax, [0, 0, 64, 149, 255, 255])
assert_equal(misc.bytescale(np.array([3, 3, 3]), low=4), [4, 4, 4])
def test_bytescale(self):
x = np.array([0, 1, 2], np.uint8)
y = np.array([0, 1, 2])
with suppress_warnings() as sup:
sup.filter(DeprecationWarning)
assert_equal(misc.bytescale(x), x)
assert_equal(misc.bytescale(y), [0, 128, 255])
def test_bytescale_keywords(self):
x = np.array([40, 60, 120, 200, 300, 500])
with suppress_warnings() as sup:
sup.filter(DeprecationWarning)
res_lowhigh = misc.bytescale(x, low=10, high=143)
assert_equal(res_lowhigh, [10, 16, 33, 56, 85, 143])
res_cmincmax = misc.bytescale(x, cmin=60, cmax=300)
assert_equal(res_cmincmax, [0, 0, 64, 149, 255, 255])
assert_equal(misc.bytescale(np.array([3, 3, 3]), low=4), [4, 4, 4])
def getProjection(self, numFrame):
""" Get projections onto yoz and xoz plane from xoy image """
xoyImg, _ = self.getGestureRegion(numFrame)
heightLen, widthLen = xoyImg.shape
depthLen = 256
size = xoyImg.size
corMap = np.zeros((size, 3))
corX = np.reshape(corMap[:, 0], (size, 1))
corY = np.reshape(corMap[:, 1], (size, 1))
corZ = np.reshape(xoyImg, (size, 1), 'F') # using Fortran-like index order
# generate x and y coordinates
for i in range(widthLen):
startIdx = i * heightLen
endIdx = startIdx + heightLen
corX[startIdx : endIdx] = np.ones((heightLen, 1)) * i
tmpArray = np.array(range(0, heightLen)) # generate matrix [0-480]
corY[startIdx : endIdx] = np.reshape(tmpArray, (tmpArray.size, 1))
# corMap[:, 0] = corX
# corMap[:, 1] = corY
# corMap[:, 2] = corZ
corMap = hstack([corX, corY, corZ])
##
thresh = 10
selectedCorMap = corMap[np.where(corMap[:, 2] > thresh)]
selectedCorMap = selectedCorMap.astype(int)
# yoz and xoz image
xozImg = np.zeros((depthLen, widthLen))
yozImg = np.zeros((heightLen, depthLen))
rowNum, _ = selectedCorMap.shape
for i in range(rowNum):
xozImg[selectedCorMap[i, 2], selectedCorMap[i, 0]] = xozImg[selectedCorMap[i, 2], selectedCorMap[i, 0]] + 1
yozImg[selectedCorMap[i, 1], selectedCorMap[i, 2]] = yozImg[selectedCorMap[i, 1], selectedCorMap[i, 2]] + 1
xozImg = bytescale(xozImg) # scale to 0-255
yozImg = bytescale(yozImg) # scale to 0-255
return xozImg, yozImg
def test_bytescale_rounding(self):
a = np.array([-0.5, 0.5, 1.5, 2.5, 3.5])
with suppress_warnings() as sup:
sup.filter(DeprecationWarning)
actual = misc.bytescale(a, cmin=0, cmax=10, low=0, high=10)
expected = [0, 1, 2, 3, 4]
assert_equal(actual, expected)
def test_bytescale_low_equals_high(self):
a = np.arange(3)
with suppress_warnings() as sup:
sup.filter(DeprecationWarning)
actual = misc.bytescale(a, low=10, high=10)
expected = [10, 10, 10]
assert_equal(actual, expected)
def test_bytescale_cscale_lowhigh(self):
a = np.arange(10)
with suppress_warnings() as sup:
sup.filter(DeprecationWarning)
actual = misc.bytescale(a, cmin=3, cmax=6, low=100, high=200)
expected = [100, 100, 100, 100, 133, 167, 200, 200, 200, 200]
assert_equal(actual, expected)
def pibayerraw(fn,exposure_sec,bit8):
with PiCamera() as cam: #load camera driver
print('camera startup gain autocal')
#LED automatically turns on, this turns it off
cam.led = False
sleep(0.75) # somewhere between 0.5..0.75 seconds to let camera settle to final gain value.
setparams(cam,exposure_sec) #wait till after sleep() so that gains settle before turning off auto
getparams(cam)
counter = 1
#%% main loop
while True:
# tic = time()
img10 = grabframe(cam)
# print('{:.1f} sec. to grab frame'.format(time()-tic))
#%% linear scale 10-bit to 8-bit
if bit8:
img = bytescale(img10,0,1024,255,0)
else:
img = img10
#%% write to PNG or JPG or whatever based on file extension
max_value = img.max()
print(max_value)
if max_value > 50:
idx = unravel_index(img.argmax(), img.shape)
xidx = idx[0]
yidx = idx[1]
print(xidx, yidx)
xlow = max(0, xidx-25)
ylow = max(0, yidx-25)
xhi = min(1944, xidx+25)
yhi = min(2592, yidx+25)
imsave(fn+'%03d' % counter + '.png',img[xlow:xhi,ylow:yhi])
counter = counter + 1
# break
return img
def make_pc_images(pca, shape):
U = REACT2D.build_dct(shape[0], shape[1], 50)
pca_images = np.empty((npca, shape[0], shape[1], 3))
pca_images[:, :, :, 0] = pca.components_[:, :ncoefs].dot(U.T[:ncoefs, :]).reshape((npca, shape[0], shape[1]))
pca_images[:, :, :, 1] = pca.components_[:, ncoefs:2*ncoefs].dot(U.T[:ncoefs, :]).reshape((npca, shape[0], shape[1]))
pca_images[:, :, :, 2] = pca.components_[:, 2*ncoefs:].dot(U.T[:ncoefs, :]).reshape((npca, shape[0], shape[1]))
npca_rows = 3
npca_cols = 3
nplots = 2
pca_idx = 0
for plot in range(nplots):
idx = 1
plt.clf()
for row in range(npca_rows):
for col in range(npca_cols):
print row, col, idx
plt.subplot(npca_rows, npca_cols, idx)
plt.imshow(bytescale(pca_images[pca_idx, :, :, :]))
plt.title('PC ' + str(pca_idx + 1))
plt.tick_params(axis='x', which='both', bottom='off', top='off', labelbottom='off')
plt.tick_params(axis='y', which='both', bottom='off', top='off', labelbottom='off')
idx += 1
pca_idx += 1
plt.savefig(plot_dir + 'PC_Images_' + str(plot + 1) + '.png')
if doshow:
plt.show()
def test_bytescale_mask(self):
a = np.ma.MaskedArray(data=[1, 2, 3], mask=[False, False, True])
actual = misc.bytescale(a)
expected = [0, 255, 3]
assert_equal(expected, actual)
assert_mask_equal(a.mask, actual.mask)
self.assertTrue(isinstance(actual, np.ma.MaskedArray))
def write_img(self,filename=None,
date=None,receiver=None,
data=None,
format='jpg',quality=80,
data_directory=None,
output_filename=None,
output_directory=None,
min_val=None,max_val=None,
verbose=True,greyscale=True,
reverse_color=True,
download_file=False,
delete_file=False,
prep=False):
"""
Write output image file containing the dynamical spectrum.
"""
if (greyscale):
mode='L'
else:
mode='RGB'
ext = format.lower()
if (data is None):
data = self.get_data(date=date,receiver=receiver,
filename=filename,
data_directory=data_directory,
download_file=download_file,
delete_file=delete_file,
verbose=verbose,prep=prep)
if (data is None):
return ""
array = data.intensity
if (min_val is None): min_val = array.min()
if (max_val is None): max_val = array.max()
array = array.clip(min_val,max_val)
if not ("(db)" in data.intensity_units.lower()):
array = to_dB(array)
array = bytescale(array)
if (reverse_color):
array = array.max() - array
image = Image.fromarray(array,mode=mode)
if (output_filename is None):
if (filename is None):
filename = self.get_filename(date,receiver=receiver)
output_filename = os.path.basename(filename)+"."+ext
if (output_directory is None):
output_path = output_filename
else:
output_path = os.path.join(output_directory,os.path.basename(output_filename))
image.save(output_path,quality=quality)
return output_path
def hdf2video(data,imgh5,outfn,clim):
outfn = Path(outfn).expanduser()
import cv2
try:
from cv2.cv import FOURCC as fourcc #Windows needs from cv2.cv
except ImportError:
from cv2 import VideoWriter_fourcc as fourcc
outfn = outfn.with_suffix('.ogv')
cc4 = fourcc(*'THEO')
# we use isColor=True because some codecs have trouble with grayscale
hv = cv2.VideoWriter(str(outfn),cc4, fps=33,
frameSize=data.shape[1:][::-1], #frameSize needs col,row
isColor=True) #right now we're only using grayscale
if not hv.isOpened():
raise TypeError('trouble starting video file')
for d in data:
#RAM usage explodes if scaling all at once on GB class file
#for d in bytescale(data,1000,4000):
#for d in sixteen2eight(data,(1000,4000)):
hv.write(gray2rgb(bytescale(d,clim[0],clim[1])))
hv.release()
def getGestureRegion(self, frameNum):
""" Get gesture region for the given frame """
# get Depth frame
depthData = self.getFrame(self.depth, frameNum)
depthGray = cv2.cvtColor(depthData, cv2.cv.CV_RGB2GRAY)
# get user segmentation frame
userSeg = self.getFrame(self.user, frameNum)
userSegGray = cv2.cvtColor(userSeg, cv2.cv.CV_RGB2GRAY)
userSegGray = cv2.medianBlur(userSegGray, 5) # Median filter on original user image
# Convert user to binary image
threshold = 128
_, userBinImg = cv2.threshold(userSegGray, threshold, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
depthGray[np.where(userBinImg == 0)] = 0
depthGray = cv2.medianBlur(depthGray, 5)
depthRealValue = depthGray.astype(np.float32) # depth value of real world (0-maxDepth)
# Convert to depth values
depthRealValue = depthRealValue / 255.0 * float(self.data['maxDepth'])
depthRealValue = depthRealValue.round()
depthRealValue = depthRealValue.astype(np.uint16)
# scale depthGray to 0-255
depthGray = depthGray.astype(np.uint16)
depthGray = bytescale(depthGray)
depthImgValue = np.copy(depthGray)
return depthImgValue, depthRealValue
def main(image):
matplotlib.rcParams["font.size"] = 10
def show_img(img, axes):
"""Plot the image as float"""
# img = img_as_float(img)
ax_img = axes
ax_img.imshow(img, cmap=plt.cm.gray)
ax_img.set_axis_off()
return ax_img
# Open and read in the fits image
try:
fits = pyfits.open(image)
# fits = Image.open(image)
except IOError:
print "Can not read the fits image: " + image + " !!"
# Check the input image
img = fits[0].data
# img = np.array(fits)
if img.ndim != 2:
raise NameError("Data need to be 2-D image !")
# Logrithm scaling of the image
img_log = np.log10(img)
img_log = img_as_float(img_log)
# Contrast streching
p5, p95 = np.percentile(img, (2, 98))
img_rescale = exposure.rescale_intensity(img, in_range=(p5, p95))
# Adaptive equalization
img_new = bytescale(img_rescale)
img_ahe = exposure.equalize_adapthist(img_new, ntiles_x=16, ntiles_y=16, clip_limit=0.05, nbins=256)
img_ahe = img_as_float(img_ahe)
# Display results
fig, axes = plt.subplots(nrows=1, ncols=3, figsize=(16, 5))
# Original image
ax_img = show_img(img_log, axes[0])
ax_img.set_title("Original")
# Contrast Enhanced one
ax_img = show_img(img_rescale, axes[1])
ax_img.set_title("Rescale")
# AHE Enhanced one
ax_img = show_img(img_ahe, axes[2])
ax_img.set_title("AHE")
# Prevent overlap of y-axis
fig.subplots_adjust(bottom=0.1, right=0.9, top=0.9, left=0.1, wspace=0.05)
# Save a PNG file
plt.gcf().savefig("ahe_test.png")
def get_galaxy_images(galaxy_id):
images = []
for c in range(3):
fname = base_dir + 'data/images_training_rev1/' + str(galaxy_id) + '_' + str(c) + '.npy'
images.append(np.load(fname))
images = np.dstack(images)
return bytescale(images) # scale the arrays for nice color representation of the image
def linear_stretch( data, min_percentile=1.0, max_percentile=97.0): pmin, pmax = numpy.percentile(data[numpy.nonzero(data)], (min_percentile, max_percentile)) data[data>pmax]=pmax data[data<pmin]=pmin bdata = misc.bytescale(data) return bdata
def save_image(filename, im, scaling='auto', depth=8):
"""Save an ndarray or image as a tiff.
Parameters
----------
im : ndarray or :class:`holopy.image.Image`
image to save.
filename : basestring
filename in which to save image. If im is an image the
function should default to the image's name field if no
filename is specified
scaling : 'auto', None, or (None|Int, None|Int)
How the image should be scaled for saving. Ignored for float
output. It defaults to auto, use the full range of the output
format. Other options are None, meaning no scaling, or a pair
of integers specifying the values which should be set to the
maximum and minimum values of the image format.
depth : 8, 16 or 'float'
What type of image to save. Options other than 8bit may not be supported
for many image types. You probably don't want to save 8bit images without
some kind of scaling.
"""
# if we don't have an extension, default to tif
if os.path.splitext(filename)[1] is '':
filename += '.tif'
# to replicate old behavior from using sp.misc.toimage
if depth == 8:
if scaling == 'auto':
cmin, cmax = None, None
else:
cmin, cmax = scaling
im = bytescale(im)
elif depth != 'float':
if scaling is not None:
if scaling == 'auto':
min = im.min()
max = im.max()
elif len(scaling) == 2:
min, max = scaling
else:
raise Error("Invalid image scaling")
if min is not None:
im = im - min
if max is not None:
im = im * ((2**depth-1)/max)
if depth == 8:
im = (im+.4999).astype('uint8')
elif depth == 16:
# PIL can't handle uint16, but seems to do the right thing
# with int16, so go ahead and use it
im = (im+.4999).astype('int16')
else:
raise Error("Unknown image depth")
PILImage.fromarray(im).save(filename, autoscale=False)
def create_composite(red, green, blue):
img_dim = red.shape
img = np.zeros((img_dim[0], img_dim[1], 3), dtype=np.float)
img[:,:,0] = red
img[:,:,1] = green
img[:,:,2] = blue
p2, p98 = np.percentile(img, (2, 98))
img_rescale = exposure.rescale_intensity(img, in_range=(0, p98))
return bytescale(img_rescale)
def test_bytescale_mask(self):
a = np.ma.MaskedArray(data=[1, 2, 3], mask=[False, False, True])
with suppress_warnings() as sup:
sup.filter(DeprecationWarning)
actual = misc.bytescale(a)
expected = [0, 255, 3]
assert_equal(expected, actual)
assert_mask_equal(a.mask, actual.mask)
assert_(isinstance(actual, np.ma.MaskedArray))
def rgb(hh, hv, spath):
hh = np.flipud(hh)
hv = np.flipud(hv)
hhhv = hh/hv
plt.figure
m.imshow(hhhv, vmin = 0., vmax = 1)
m.drawrivers(color = 'black', linewidth = 1.)
m.drawstates(color = 'black', linewidth = 1.5)
m.drawparallels(parallels, linewidth = 0.0, labels = [1,0,0,0], fontsize=10)
m.drawmeridians(meridians, linewidth = 0.0, labels = [0,0,0,1], fontsize=10)
cbar = m.colorbar()
cbar.set_label('')
title0 = 'ALOS-1 HH/HV'
plt.title(title0)
plt.savefig(spath + 'hhhv.png', transparent = False, bbox_inches = 'tight', dpi = 300)
plt.close()
r = bytescale(hh, cmin = -30, cmax = -10)
g = bytescale(hh, cmin = -30, cmax = -10)
b = bytescale(hhhv, cmin = 0.5, cmax = 1)
rgb1 = np.zeros( (rows,cols,3), 'uint8' )
rgb1[...,0] = r
rgb1[...,1] = g
rgb1[...,2] = b
plt.figure
m.imshow(rgb1)
m.drawrivers(color = 'black', linewidth = 1.)
m.drawstates(color = 'black', linewidth = 1.5)
m.drawparallels(parallels, linewidth = 0.0, labels = [1,0,0,0], fontsize=10)
m.drawmeridians(meridians, linewidth = 0.0, labels = [0,0,0,1], fontsize=10)
#cbar = m.colorbar()
#cbar.set_label('')
title0 = 'ALOS-1 RGB'
plt.title(title0)
plt.savefig(spath + 'rgb1.png', transparent = False, bbox_inches = 'tight', dpi = 300)
plt.close()
return hhhv
def get_array(self, band=1):
"""
Get a band as a 32-bit numpy array
Parameters
----------
band : int
The band to read, default 1
"""
array = self.geodata.read_array(band=band)
return bytescale(array)
def sdo_aia_scale(image=None, wavelength=None):
'''
rescale the aia image
:param image: normalised aia image data
:param wavelength:
:return: byte scaled image data
'''
from scipy.misc import bytescale
clrange = sdo_aia_scale_dict(wavelength)
image[image > clrange['high']] = clrange['high']
image[image < clrange['low']] = clrange['low']
image = np.log10(image)
return bytescale(image)
def data_to_bytescale_rgb(
data): # used to create the SOURCE PNGs (MRI, FA, MD)
"""
Converts a single-channel grayscale image to a 3-channel image that can be
then saved as a PNG
"""
im = bytescale(data)
w, h = im.shape
ret = np.empty((w, h, 3), dtype=np.uint8)
ret[:, :, 0] = im
ret[:, :, 1] = im
ret[:, :, 2] = im
return ret
def get_geodataset(self, nodeIndex):
"""
Constructs a GeoDataset object from the given node image and assigns the
dataset and its NumPy array to the 'handle' and 'image' node attributes.
Parameters
----------
nodeIndex : int
The index of the node.
"""
self.node[nodeIndex]['handle'] = GeoDataset(self.node[nodeIndex]['image_path'])
self.node[nodeIndex]['image'] = bytescale(self.node[nodeIndex]['handle'].read_array())
def __init__(self, paths, fn, fn_mapping, has_alpha, scale_factor=1.0):
super().__init__(paths,
fn,
fn_mapping,
has_alpha,
scale_factor=scale_factor)
fpath = os.path.join(self.paths['images'],
self.fn_mapping['images'](self.fn))
_, fname = os.path.split(fpath)
if "*" in fname:
files = glob.glob(fpath)
assert len(
files
) == 1, 'Multiple files match the image file name pattern, please use a unique pattern.'
fpath = files[0]
lfpath = fpath.lower()
if not lfpath.endswith('.jpg') and not lfpath.endswith(
'.jpeg') and not lfpath.endswith(
'.png') and not lfpath.endswith(
'.tif') and not lfpath.endswith(
'.tiff') and not lfpath.endswith(
'.bmp') and not lfpath.endswith('.gif'):
raise Exception('Unsupported file format: ' + fname)
if fpath.lower().endswith('.tif') or fpath.lower().endswith('.tiff'):
img = imread(fpath)
else:
img = imread(fpath, pilmode="RGB")
if len(img.shape) == 2:
rgb_img = np.stack((img, ) * 3, axis=-1)
else:
assert len(img.shape) == 3
# has alpha channel?
if img.shape[2] == 4:
rgb_img = img[:, :, :-1]
else:
rgb_img = img
self.im = bytescale(rgb_img)
if '646f5e00a2db3add97fb80a83ef3c07edd1b17b1b0d47c2bd650cdcab9f322c0' in fn:
self.im = cv2.imread(os.path.join(self.paths['images'], self.fn),
cv2.IMREAD_COLOR)
if scale_factor != 1.0:
width, height, _ = self.im.shape
self.im = imresize(
self.im,
(int(scale_factor * width), int(scale_factor * height)),
interp='bicubic')
def linear_stretch(data, min_percentile=1.0, max_percentile=97.0): if verbose: print 'linear_stretch', numpy.min(data), numpy.mean(data), numpy.max(data), min_percentile, max_percentile pmin, pmax = numpy.percentile(data[numpy.nonzero(data)], (min_percentile, max_percentile)) if verbose: print "pmin:", pmin print "pmax:", pmax data[data>pmax]=pmax data[data<pmin]=pmin bdata = misc.bytescale(data) return bdata
def check():
#imverting the image
img = misc.imread('assets/out.png')
img = (255 - cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)) / 255
#resizing image into 8X8 matrix
img = misc.imresize(img, (28, 28))
img = img.astype('float32')
#changing byte scale from 16 to 0 according to imput data
img = misc.bytescale(img, high=16, low=0)
flat_img = img.reshape(1, 784)
result = model.predict_classes(flat_img)
return result
def pc_image(pca, shape, pc_idx):
U = REACT2D.build_dct(shape[0], shape[1], 50)
pca_images = np.empty((shape[0], shape[1], 3))
pca_images[:, :, 0] = pca.components_[pc_idx, :ncoefs].dot(U.T[:ncoefs, :]).reshape((shape[0], shape[1]))
pca_images[:, :, 1] = pca.components_[pc_idx, ncoefs:2*ncoefs].dot(U.T[:ncoefs, :]).reshape((shape[0], shape[1]))
pca_images[:, :, 2] = pca.components_[pc_idx, 2*ncoefs:].dot(U.T[:ncoefs, :]).reshape((shape[0], shape[1]))
plt.imshow(bytescale(pca_images))
plt.title('PC ' + str(pc_idx + 1))
plt.tick_params(axis='x', which='both', bottom='off', top='off', labelbottom='off')
plt.tick_params(axis='y', which='both', bottom='off', top='off', labelbottom='off')
plt.show()
def loadData(self, list_file_names):
x_train = []
y_labels = [1, 2, 3, 4, 5, 6, 7, 8, 9, 1, 0, 0, 0, 2, 2, 2]
for name in list_file_names:
img = misc.imread(name)
img = misc.imresize(img, (8, 8))
img = img.astype(np.float64)
img = misc.bytescale(img, high=16, low=0)
x_test = []
for row in img:
for pixel in row:
x_test.append(sum(pixel) / 3.0)
x_train.append(x_test)
return x_train, y_labels
def linear_stretch(data, min_percentile=1.0, max_percentile=97.0):
if verbose:
print 'linear_stretch', numpy.min(data), numpy.mean(data), numpy.max(
data), min_percentile, max_percentile
pmin, pmax = numpy.percentile(data[numpy.nonzero(data)],
(min_percentile, max_percentile))
if verbose:
print "pmin:", pmin
print "pmax:", pmax
data[data > pmax] = pmax
data[data < pmin] = pmin
bdata = misc.bytescale(data)
return bdata
def img2array(image, type):
# Reading image
img = misc.imread(image)
# Resizing image to 8 * 8 = 64
img = misc.imresize(img, (8, 8))
# Updating type from unsigned int 8 to signed float 64
img = img.astype(type)
# We have data set image of integer pixels in the range 0..16.
img = misc.bytescale(img, high=16, low=0)
array = []
# Converting out 3D array to 1D Array
for each_row in img:
for each_pixel in each_row:
array.append(((sum(each_pixel)) / 3.0))
return array
def find_Combination(drug1, drug2, batch):
data = pandas.read_sql(
"select Image_FileName_DAPI,Image_FileName_Mitotracker,Image_FileName_BetaTubulin,Image_PathName_DAPI from DPN1018"
+ batch + "Per_Image where (Image_Metadata_ID_A = '" + drug1 +
"' and Image_Metadata_ID_B = '" + drug2 +
"') or (Image_Metadata_ID_A = '" + drug2 +
"' and Image_Metadata_ID_B = '" + drug1 + "');",
con=db)
if len(data) == 0:
print 'No Images found (Check spelling)'
exit()
images = []
foci = []
for image, image2, image3, path in zip(data['Image_FileName_DAPI'],
data['Image_FileName_BetaTubulin'],
data['Image_FileName_Mitotracker'],
data['Image_PathName_DAPI']):
path = path.split('lab_menche')[1]
foci.append(int(image.split('f0')[1].split('p')[0]))
k = misc.imread("/Volumes/scratch/lab_menche" + path + "/" + image)
k2 = misc.imread("/Volumes/scratch/lab_menche" + path + "/" + image2)
k3 = misc.imread("/Volumes/scratch/lab_menche" + path + "/" + image3)
#print k
#exit()
rgb = np.dstack((k3, k2, k))
#r_rgb = misc.imresize(rgb, 100)
#plt.imshow(rgb)
#plt.show()
test = misc.bytescale(rgb)
#plt.imshow(test)
#plt.show()
#plt.close()
images.append(test)
return create_combined_Image(images, foci)
def calculate_color_features(data):
""" Calculates color features related to the greenness of each point.
The default features are [a, b, NGRDVI] where a and b are the green-red and
blue-yellow coordinates of the CIE-Lab color space.
Args:
data: An n x d array of input data. Rows are [x, y, z, r, g, b, ...]
Returns:
An n x 3 array of features for each point.
"""
rgb = bytescale(data[:, 3:6]).astype(np.int16)
lab = rgb2lab(np.array([rgb]))[0].reshape(-1, 3)
ngrdvi = calculate_ngrdvi(data).reshape(-1, 1)
return np.hstack([lab[:, 1:3], ngrdvi])
def nhance(frames, pct=1.0, frs=60, swp=30):
with tf.Session(config=config) as sess:
x_, enhanced = build_net()
saver = tf.train.Saver()
saver.restore(sess, "models/iphone")
imgs = np.float16(frames) / 255
vlen = frames.shape[0]
ret = np.empty(frames.shape, dtype=np.uint8)
if pct < 1.0:
ret_ba = np.copy(frames)
lim_pt = int(img_w * (1.0 - pct))
else:
ret_ba = None
ret_sw = np.copy(frames)
ptr = 0
while ptr < vlen:
print('Ptr is: ', ptr)
bsz = batch_sz if ptr + batch_sz <= vlen else (vlen - ptr)
batch = np.reshape(imgs[ptr:ptr + bsz], [bsz, img_sz])
r_batch = sess.run(enhanced, feed_dict={x_: batch})
r_batch = bytescale(np.reshape(r_batch,
[bsz, img_h, img_w, img_d]))
ret[ptr:ptr + batch_sz, :, :, :] = r_batch
if pct < 1.0:
ret_ba[ptr:ptr + batch_sz, :, lim_pt:, :] = r_batch[:, :,
lim_pt:, :]
if ptr > frs + swp:
ret_sw[ptr:ptr + batch_sz, :, :, :] = r_batch
else:
for j in range(bsz):
if ptr + j <= frs:
continue
elif ptr + j <= frs + swp:
prop = ((ptr + j - frs) * 1.0) / (swp * 1.0)
ppoz = int(img_w * prop)
ret_sw[ptr + j, :, :ppoz, :] = r_batch[j, :, :ppoz, :]
else:
ret_sw[ptr + j, :, :, :] = r_batch[j, :, :, :]
ptr += bsz
return ret, ret_ba, ret_sw
def data_to_labels_rgb(data, NUM_LABELS, RANGE_MIN, RANGE_MAX):
"""
Rescales the histology for deep learning processing.
Since labels start at zero, the high parmeter is NUM_LABELS-1
the cmin and cmax parameters depend on the properties of the histology
and can be changed to obtain a different linear mapping of the final labels
Example:
test = np.linspace(0,1,num=100)
array([ 0. , 0.01010101, 0.02020202, 0.03030303, 0.04040404,
0.05050505, 0.06060606, 0.07070707, 0.08080808, 0.09090909,
...
0.95959596, 0.96969697, 0.97979798, 0.98989899, 1. ])
bytescale(test,low=0, high=15, cmin=0,cmax=0.5)
array([ 0, 0, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5,
5, 5, 6, 6, 6, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9, 10, 10,
10, 11, 11, 11, 12, 12, 12, 12, 13, 13, 13, 14, 14, 14, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15], dtype=uint8)
bytescale(test,low=0, high=15-1, cmin=0,cmax=0.5)
array([ 0, 0, 1, 1, 1, 1, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 5,
5, 5, 5, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8, 9, 9, 9,
10, 10, 10, 10, 11, 11, 11, 12, 12, 12, 12, 13, 13, 13, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14], dtype=uint8)
"""
im = bytescale(data,
low=0,
high=NUM_LABELS - 1,
cmin=RANGE_MIN,
cmax=RANGE_MAX)
w, h = data.shape
ret = np.empty((w, h, 3), dtype=np.uint8)
ret[:, :, 0] = im
ret[:, :, 1] = im
ret[:, :, 2] = im
return ret
def load_data(fldr):
IMAGE_SIZE = 32
images, labels, uids = [], [], []
width, height = IMAGE_SIZE, IMAGE_SIZE
MAX_NUM_DOMAINS = 110
uid = 0
cache_fname = 'data/lipitk.pkl'
if os.path.exists(cache_fname):
images, labels, uids = pickle.load(open(cache_fname, "rb"))
else:
for dom in tqdm.tqdm(range(MAX_NUM_DOMAINS)):
dom_fldr = "%s/usr_%d" % (fldr, dom)
if not os.path.exists(dom_fldr):
continue
for fname in os.listdir(dom_fldr):
if fname.find('.tiff') < 0:
continue
li = int(fname.split('t')[0])
img = misc.imread(dom_fldr + "/" + fname)
img = misc.imresize(img, (height, width))
img = img.astype(np.float32)
img = misc.bytescale(img)
img = img.astype(np.uint8)
assert np.max(img) <= 255 and np.min(
img) >= 0, "Max and min of image: %f %f" % (np.max(img),
np.min(img))
img = (img - 128.) / 128.
assert np.max(img) != np.min(img)
images.append(img)
labels.append(li)
uids.append(uid)
uid += 1
pickle.dump((images, labels, uids), open(cache_fname, "wb"))
print("Labels: %s uids: %s" % (labels[:10], uids[:10]))
print("Labels: %s uids: %s" % (labels[-10:], uids[-10:]))
print("Test images: ", np.max(images[0]), np.min(images[0]))
print("Read %d examples" % len(images))
uids = np.array(uids)
np.random.shuffle(uids)
return np.array(images), np.array(labels), uids
def _add_to_tfrecord(image_dir, tfrecord_writer, split_name):
"""Loads images and writes files to a TFRecord.
Args:
image_dir: The image directory where the raw images are stored.
list_filename: The list file of images.
tfrecord_writer: The TFRecord writer to use for writing.
"""
# filenames = tf.train.match_filenames_once(os.path.join(image_dir,'\*.jpg'))
list = os.listdir(image_dir)
i = 0
count = 0
for filename in list:
i += 1
filenames = os.listdir(os.path.join(image_dir + "/", filename))
nums = len(filenames)
count += nums
shape = (_IMAGE_HEIGHT, _IMAGE_WIDTH, _IMAGE_CHANNELS)
with tf.Graph().as_default():
image = tf.placeholder(dtype=tf.uint8, shape=shape)
encoded_png = tf.image.encode_jpeg(image)
j = 0
with tf.Session('') as sess:
for line in filenames:
sys.stdout.write(
'\r>> Converting %s%s image %d/%d now%d images' %
(filename, split_name, j + 1, nums, count))
sys.stdout.flush()
j += 1
image_data = misc.imread(
os.path.join(image_dir + "/" + filename + "/", line))
label = i - 1
image_data = misc.bytescale(image_data)
image_data = misc.imresize(image_data,
[_IMAGE_HEIGHT, _IMAGE_WIDTH])
png_string = sess.run(encoded_png,
feed_dict={image: image_data})
example = image_to_tfexample(png_string, label,
bytes(line,
'utf-8'), _IMAGE_HEIGHT,
_IMAGE_WIDTH, b'jpg')
tfrecord_writer.write(example.SerializeToString())
def process(source, IMAGE_SIZE=224):
ds = dicom.read_file(source)
pixel_array = ds.pixel_array
height, width = pixel_array.shape
if height < width:
pixel_array = pixel_array[:,
int((width - height) /
2):int((width + height) / 2)]
else:
pixel_array = pixel_array[int((height - width) /
2):int((width + height) / 2), :]
im = cv2.resize(pixel_array, (IMAGE_SIZE, IMAGE_SIZE))
im = bytescale(im)
# im = im / 256
im = np.dstack((im, im, im))
im = im[:, :, [2, 1, 0]]
im = im.transpose((2, 0, 1))
return im
def check():
img = misc.imread('assets/out.png')
#imverting the image
img = 255 - cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
#resizing image into 8X8 matrix
img = misc.imresize(img, (8, 8))
img = img.astype('float64')
#changing byte scale from 16 to 0 according to imput data
img = misc.bytescale(img, high=16, low=0)
# uncomment if you want to see the image
# plt.imshow(img, cmap=plt.cm.gray_r, interpolation='nearest')
flat_img = img.reshape(1, 64)
result = clf.predict(flat_img)
return result
def imshow(image, colormap=False, video=False):
import imageio
import iterm2_tools
from matplotlib import cm
from scipy.misc import bytescale
from PIL import Image
from iterm2_tools.images import display_image_bytes
if type(image).__name__ == 'Variable':
image = image.data
if 'torch.cuda' in type(image).__module__:
image = image.cpu()
if 'Tensor' in type(image).__name__:
image = image.numpy()
if colormap:
image = (cm.Blues(image) * 255).astype(np.uint8)
else:
image = bytescale(image)
if image.ndim == 4:
video = True
if image.ndim==3 and (image.shape[0] not in [1,3] and image.shape[-1] not in [1,3]):
video = True
if video:
if image.shape[1] == 3:
image = image.transpose([2,3,1]).astype(np.uint8)
image = image.squeeze()
if image.ndim == 2:
image = image[None]
images = [im for im in image]
s = imageio.mimsave(imageio.RETURN_BYTES, images, format='gif', duration=0.3)
print(display_image_bytes(s))
else:
if image.shape[0] == 3:
image = image.transpose([1,2,0]).astype(np.uint8)
image = image.squeeze()
s = imageio.imsave(imageio.RETURN_BYTES, image, format='png')
s = display_image_bytes(s)
# Depending on the version of iterm2_tools, display_image_bytes can
# either print directly to stdout or return the string to print.
if s is not None:
print(s)
def _create_blanks(blank_sec_length, on_msec_length, off_msec_length,
stimulus_shape, blank_loc):
"""create the blank stimuli
stimulus_shape: 2-tuple of ints. specifies the shape of a single stimulus.
"""
nblanks = blank_sec_length / ((on_msec_length + off_msec_length) / 1000.)
if nblanks != int(nblanks):
raise Exception(
"Because of your timing ({loc}_blank_sec_length, on_msec_length, and "
"off_msec_length), I can't show blanks for the {loc} {length:.02f} seconds"
". {loc}_blank_sec_length must be a multiple of on_msec_length+"
"off_msec_length!".format(loc=blank_loc, length=blank_sec_length))
nblanks = int(nblanks)
blanks = smisc.bytescale(np.zeros(
(nblanks, stimulus_shape[0], stimulus_shape[1])),
cmin=-1,
cmax=1)
return nblanks, blanks
def main():
fig = plt.figure()
axis = [fig.add_subplot(211 + _) for _ in range(2)]
s = open(XPM_INFILE, 'rb').read()
nda = xpm.XPM(s) # as ndarray (dtype=np.uint8) BGR(A)
sys.stderr.write('%s\n' % str(nda))
r, c, m = nda.shape
img = Image.frombuffer('RGBA', (c, r), nda, 'raw', 'BGRA', 0, 1)
img.show() # PIL.Image
bm = np.array(img) # RGB(A)
axis[0].imshow(bm)
# misc.imsave(XPM_OUTGIF_0, np.uint8(bm)) # changed
misc.imsave(XPM_OUTGIF_0, np.float32(bm)) # color changed
# misc.imsave(XPM_OUTGIF_1, misc.bytescale(bm, cmin=0, cmax=255)) # changed
misc.imsave(XPM_OUTGIF_1, misc.bytescale(bm)) # color changed
# im = misc.toimage(bm, cmin=0, cmax=255) # same as PIL.Image
im = misc.toimage(bm) # same as PIL.Image
im.save(XPM_OUTPNG) # palette ok
axis[1].imshow(im)
plt.show()
def visualize_intermidiate_output(model, X, layer_number, folder_path,
visualize_train_or_test):
get_layer_output = K.function(
[model.layers[0].input, K.learning_phase()],
[model.layers[layer_number].output])
if visualize_train_or_test == 'train':
layer_output = get_layer_output([X, 1])[0]
if visualize_train_or_test == 'test':
layer_output = get_layer_output([X, 0])[0]
print(layer_number, layer_output.shape)
if not os.path.exists(folder_path + str(layer_number) + '/'):
os.makedirs(folder_path + str(layer_number) + '/')
count = 1
for data in layer_output[0]:
img_i = bytescale(data)
pic_name = folder_path + str(layer_number) + '/' + str(count) + '.jpg'
cv2.imwrite(pic_name, img_i)
count += 1
def proc_imgs_comp(i, warp_matrices, bndFolders, panel_irradiance):
i.compute_reflectance(panel_irradiance)
#i.plot_undistorted_reflectance(panel_irradiance)
cropped_dimensions, edges = imageutils.find_crop_bounds(i, warp_matrices)
im_aligned = imageutils.aligned_capture(i, warp_matrices,
cv2.MOTION_HOMOGRAPHY,
cropped_dimensions,
None, img_type="reflectance")
im_display = np.zeros((im_aligned.shape[0],im_aligned.shape[1],5), dtype=np.float32 )
for iM in range(0,im_aligned.shape[2]):
im_display[:,:,iM] = imageutils.normalize(im_aligned[:,:,iM])
rgb = im_display[:,:,[2,1,0]]
#cir = im_display[:,:,[3,2,1]]
RRENir = im_display[:,:,[4,3,2]]
imoot = [rgb, RRENir]
imtags = ["RGB.tif", "RRENir.tif"]
im = i.images[1]
hd, nm = os.path.split(im.path[:-5])
for ind, k in enumerate(bndFolders):
img8 = bytescale(imoot[ind])
outfile = os.path.join(k, nm+imtags[ind])
imageio.imwrite(outfile, img8)
cmd = ["exiftool", "-tagsFromFile", im.path, "-file:all", "-iptc:all",
"-exif:all", "-xmp", "-Composite:all", outfile,
"-overwrite_original"]
call(cmd)
def auto_contrast(image, low=0.02, high=0.99):
max_val = np.max(image)
min_val = np.min(image)
imb = misc.bytescale(image)
xh = np.array(range(257))
histo = np.histogram(imb, bins=xh)
xh = histo[1][0:-1]
yh = histo[0]
# plot(xh,yh)
# show()
yhtot = np.sum(yh)
# Get mininmum level
yh_i = 0.0
i = 0
if (low <= 0.0):
lev_min = min_val
else:
while (yh_i < low * yhtot):
yh_i += yh[i]
i += 1
lev_min = (max_val - min_val) * (float(xh[i - 1]) / 255.0) + min_val
# Get maximum level
yh_i = 0.0
i = 0
if (high >= 1.0):
lev_max = max_val
else:
while (yh_i < high * yhtot):
yh_i += yh[i]
i += 1
lev_max = (max_val - min_val) * (float(xh[i - 1]) / 255.0) + min_val
il = np.where(image <= lev_min)
image[il] = lev_min
ih = np.where(image >= lev_max)
image[ih] = lev_max
return image
def pibayerraw(fn, exposure_sec, bit8):
with PiCamera() as cam: #load camera driver
print('camera startup gain autocal')
#LED automatically turns on, this turns it off
cam.led = False
sleep(
0.75
) # somewhere between 0.5..0.75 seconds to let camera settle to final gain value.
setparams(
cam, exposure_sec
) #wait till after sleep() so that gains settle before turning off auto
getparams(cam)
counter = 1
#%% main loop
while True:
# tic = time()
img10 = grabframe(cam)
# print('{:.1f} sec. to grab frame'.format(time()-tic))
#%% linear scale 10-bit to 8-bit
if bit8:
img = bytescale(img10, 0, 1024, 255, 0)
else:
img = img10
#%% write to PNG or JPG or whatever based on file extension
max_value = img.max()
print(max_value)
if max_value > 50:
idx = unravel_index(img.argmax(), img.shape)
xidx = idx[0]
yidx = idx[1]
print(xidx, yidx)
xlow = max(0, xidx - 25)
ylow = max(0, yidx - 25)
xhi = min(1944, xidx + 25)
yhi = min(2592, yidx + 25)
imsave(fn + '%03d' % counter + '.png', img[xlow:xhi, ylow:yhi])
counter = counter + 1
# break
return img
def get_array(self, nodeindex, downsampling=1):
"""
Downsample the input image file by some amount using bicubic interpolation
in order to reduce data sizes for visualization and analysis, e.g. feature detection
Parameters
----------
nodeindex : hashable
The index into the node containing a geodataset object
downsampling : int
[1, infinity] downsampling
"""
array = self.node[nodeindex]['handle'].read_array()
newx_size = int(array.shape[0] / downsampling)
newy_size = int(array.shape[1] / downsampling)
resized_array = imresize(array, (newx_size, newy_size), interp='bicubic')
self.node[nodeindex]['image'] = bytescale(resized_array)
self.node[nodeindex]['image_downsampling'] = downsampling
def hough_circle_image(img, dp, mindist, param1, param2, minr, maxr):
img = bytescale(img)
circles = cv2.HoughCircles(img,
cv2.HOUGH_GRADIENT,
dp,
mindist,
param1=param1,
param2=param2,
minRadius=minr,
maxRadius=maxr)
if circles is not None:
circles = np.uint16(np.around(circles))
for i in circles[0, :]:
# draw the outer circle
cv2.circle(img, (i[0], i[1]), i[2], (0, 255, 0), 2)
# draw the center of the circle
cv2.circle(img, (i[0], i[1]), 2, (0, 0, 255), 3)
return (img, circles)
def main(image):
#global path
#a = 0
#stop = False
#while not stop:
#if a < len(commands):
#for n, i in enumerate(commands):
#i = path + i
digits = datasets.load_digits()
n_samples = len(digits.images)
#features = digits.data
features = digits.images.reshape((n_samples, -1))
labels = digits.target
clf = SVC(gamma=0.0001, C=100)
clf.fit(features[:n_samples], labels[:n_samples])
img = imread(image)
#print(f"{img}img = imread(i)")
img = transform.resize(img, (8, 8))
#print(f"{img}img = transform.resize(img, (8,8))")
img = img.astype(digits.images.dtype)
#img = transform.resize(img, (8,8))
#print(f"{img}img = img.astype(digits.images.dtype)")
img = bytescale(img, high=16, low=0)
imshow(img)
#print(f"{img}img = bytescale(img, high=16, low=0)")
x_test = []
for eachRow in img:
for eachPixel in eachRow:
x_test.append(sum(eachPixel) / 3)
clf.predict([x_test])
#print('\n' * 2)
#print(f"{clf.predict([x_test])} - recognized digit :)")
#print('\n' * 2)
#a += 1
save_data(clf.predict([x_test])) #
def auto_contrast(image,low=0.02,high=0.99):
max_val = np.max(image)
min_val = np.min(image)
imb = misc.bytescale(image)
xh = np.array(range(257))
histo = np.histogram(imb,bins=xh)
xh = histo[1][0:-1]
yh = histo[0]
# plot(xh,yh)
# show()
yhtot = np.sum(yh)
# Get mininmum level
yh_i = 0.0 ; i=0
if (low <= 0.0):
lev_min = min_val
else:
while (yh_i < low*yhtot):
yh_i += yh[i]
i += 1
lev_min = (max_val - min_val)*(float(xh[i-1])/255.0) + min_val
# Get maximum level
yh_i = 0.0 ; i=0
if (high >= 1.0):
lev_max = max_val
else:
while (yh_i < high*yhtot):
yh_i += yh[i]
i += 1
lev_max = (max_val - min_val)*(float(xh[i-1])/255.0) + min_val
il = np.where(image <= lev_min)
image[il] = lev_min
ih = np.where(image >= lev_max)
image[ih] = lev_max
return image
def nodules_detection(path):
if len(os.listdir('/output/Detector_output/')) == 0:
detector.detect_nodules('/output/DataPreprocessing_output/',
'/output/Detector_output/')
npy_files = os.listdir('/output/Detector_output/')
if not os.path.exists('/output/Detector_images/'):
os.makedirs('/output/Detector_images/')
if len(npy_files) > 0:
filenames = []
for file in npy_files:
arr = np.load(os.path.join('/output/Detector_output/', file))
for i, scan in enumerate(arr):
scan = bytescale(scan)
img = Image.fromarray(scan)
filename = file.replace('.npy', str(i)) + '.jpg'
filenames.append(filename)
img.save('/output/Detector_images/' + filename)
#json_arr = jsonify(filenames = filenames)
return render_template('nodules.html', filenames=filenames)
else:
return "No scans to detect Nodules.."
def recognize():
digits = datasets.load_digits()
features = digits.data
labels = digits.target
clf = SVC(gamma=0.001)
clf.fit(features, labels)
for i in range(1, 8):
img = misc.imread('c_' + str(i) + ".png")
# img = misc.imresize(img, (8,8))
img = img.astype(digits.images.dtype)
print(img.dtype)
img = misc.bytescale(img, high=16, low=0)
x_test = []
for eachRow in img:
# for eachPixel in eachRow:
x_test.append(sum(eachRow) / 3.0)
print(clf.predict([x_test]))
def image_save(fname, img_array, low=-1, high=1):
"""
Saves the image ``img_array`` to ``fname``.
Args:
fname: Save image under this filename
img_array: array of shape ``(height, width)``, ``(1, height, width)``,
or ``(3, height, width)``.
low: Lowest pixel value
high: Highest pixel value.
The image range must be between ``low`` and ``high``.
"""
if img_array.dtype != np.uint8:
if img_array.min() < low:
raise Exception("Got an image with min {}, but low is set to {}."
.format(img_array.min(), low))
if img_array.max() > high:
raise Exception("Got an image with max {}, but high is set to {}."
.format(img_array.min(), low))
ndim = len(img_array.shape)
nb_channels = len(img_array)
if ndim == 3 and nb_channels == 1:
img = img_array[0]
elif ndim == 3 and nb_channels == 3:
img = np.moveaxis(img_array, 0, -1)
elif ndim == 2:
img = img_array
else:
raise Exception("Did not understand image shape: {}".format(img_array.shape))
if img_array.dtype == np.uint8:
img_bytes = img_array
else:
img_0_to_1 = (img - low) / (high - low)
img_bytes = bytescale(img_0_to_1 * 255, cmin=0, cmax=255)
toimage(img_bytes).save(fname)
def make_cca_images(cca, shape, dct_idx=None):
n_components = cca.x_weights_.shape[1]
U = REACT2D.build_dct(shape[0], shape[1], 50)
dct_idx = np.arange(2499)
U = U[:, dct_idx]
cca_images = np.empty((n_components, shape[0], shape[1], 3))
cca_images[:, :, :, 0] = \
cca.components_[:, dct_idx].dot(U.T).reshape((n_components, shape[0], shape[1]))
cca_images[:, :, :, 1] = \
cca.components_[:, dct_idx + len(dct_idx)].dot(U.T).reshape((n_components, shape[0], shape[1]))
cca_images[:, :, :, 2] = \
cca.components_[:, dct_idx + 2*len(dct_idx)].dot(U.T).reshape((n_components, shape[0], shape[1]))
ncca_rows = 3
ncca_cols = 3
nplots = 2
cca_idx = 0
for plot in range(nplots):
idx = 1
plt.clf()
for row in range(ncca_rows):
for col in range(ncca_cols):
print row, col, idx
plt.subplot(ncca_rows, ncca_cols, idx)
plt.imshow(bytescale(cca_images[cca_idx, :, :, :]))
plt.title('CCA ' + str(cca_idx + 1))
plt.tick_params(axis='x', which='both', bottom='off', top='off', labelbottom='off')
plt.tick_params(axis='y', which='both', bottom='off', top='off', labelbottom='off')
idx += 1
cca_idx += 1
plt.savefig(plot_dir + 'CCA_Images_' + str(plot + 1) + '.png')
if doshow:
plt.show()
def extract_features(image_array, extractor_parameters):
"""
This method finds and extracts features from an image using the given dictionary of keyword arguments.
The input image is represented as NumPy array and the output features are represented as keypoint IDs
with corresponding descriptors.
Parameters
----------
image_array : ndarray
a NumPy array that represents an image
extractor_parameters : dict
A dictionary containing OpenCV SIFT parameters names and values.
Returns
-------
: tuple
in the form ([list of OpenCV KeyPoints], [NumPy array of descriptors as geometric vectors])
"""
sift = cv2.xfeatures2d.SIFT_create(**extractor_parameters)
converted_array = misc.bytescale(image_array)
return sift.detectAndCompute(converted_array, None)
def _load_lipitk(fldr):
images, labels, uids = [], [], []
IMAGE_SIZE = 32
width, height = IMAGE_SIZE, IMAGE_SIZE
MAX_NUM_DOMAINS = 110
uid = 0
for dom in range(MAX_NUM_DOMAINS):
dom_fldr = "%s/usr_%d" % (fldr, dom)
if not os.path.exists(dom_fldr):
continue
for fname in os.listdir(dom_fldr):
if fname.find('.tiff') < 0:
continue
li = int(fname.split('t')[0])
img = misc.imread(dom_fldr + "/" + fname)
img = misc.imresize(img, (height, width))
img = img.astype(np.float32)
img = misc.bytescale(img)
img = img.astype(np.uint8)
assert np.max(img) <= 255 and np.min(
img) >= 0, "Max and min of image: %f %f" % (np.max(img),
np.min(img))
img = img / 255.
images.append(img)
labels.append(li)
uids.append(uid)
uid += 1
print("Labels: %s uids: %s" % (labels[:10], uids[:10]))
print("Labels: %s uids: %s" % (labels[-10:], uids[-10:]))
print("Test images: ", np.max(images[0]), np.min(images[0]))
print("Read %d examples" % len(images))
return np.array(images), np.array(labels), np.array(uids)
def ScaleImage( Image ):
'''
Create a byte scaled image with specified histogram
'''
#Range = numpy.array([0.0, 0.8])
Range = numpy.array([0.0, 1.0])
ScaledImage = bytescale( Image, cmin=Range[0], cmax=Range[1] )
# MODIS Rapid Response Enhancement for True Colour
# x array of input values
x = numpy.array([0, 30, 60, 120, 190, 255])
# y array of output values
y = numpy.array([0, 110, 160, 210, 240, 255])
# Create output array
rows = Image.shape[0]
cols = Image.shape[1]
Scaled = numpy.zeros( (rows,cols), numpy.uint8 )
for i in range(x.shape[0] - 1):
x1 = x[i]
x2 = x[i + 1]
y1 = y[i]
y2 = y[i + 1]
m = (y2 - y1) / float((x2 - x1))
b = y2 - (m * x2)
mask = numpy.where( (ScaledImage >= x1) & (ScaledImage < x2) )
Scaled[mask] = (m * ScaledImage + b)[mask]
mask = numpy.where( ScaledImage >= x2 )
Scaled[mask] = 255
return Scaled
def showImage(self):
#https://github.com/shuge/Enjoy-Qt-Python-Binding/blob/master/image/display_img/pil_to_qpixmap.py
#myQtImage = ImageQt(im)
#qimage = QtGui.QImage(myQtImage)
im = self.image
if (self.divide_background.isChecked()
and self.background_image is not None):
im = np.true_divide(im, self.background_image)
im = bytescale(im)
elif self.bit_depth > 8:
# if we ask the camera for more than 8 bits, we will get a 16 bit
# image that uses the upper bits, so discard the lower 8 bits to get
# something we can show on the screen
im = im / 2**8
im = to_pil_image(im)
data = im.convert("RGBA").tostring('raw', "RGBA")
qim = QtGui.QImage(data, self.roi_shape[0], self.roi_shape[1],
QtGui.QImage.Format_ARGB32)
pixmap = QtGui.QPixmap.fromImage(qim)
myScaledPixmap = pixmap.scaled(QtCore.QSize(900, 900))
self.frame.setPixmap(myScaledPixmap)
def GuessAnchorRegion(whole_img, sample_region):
"""
It detects a region with clean edges, proper for drift measurements. This region
must not overlap with the sample that is to be scanned due to the danger of
contamination.
whole_img (ndarray): 2d array with the whole SEM image
sample_region (tuple of 4 floats): roi of the sample in order to avoid overlap
returns (tuple of 4 floats): roi of the anchor region
"""
# Drift correction region shape
dc_shape = (50, 50)
# Properly modified image for cv2.Canny
uint8_img = misc.bytescale(whole_img)
# Generates black/white image that contains only the edges
cannied_img = cv2.Canny(uint8_img, 100, 200)
# Mask the sample_region plus a margin equal to the half of dc region and
# a margin along the edges of the whole image again equal to the half of
# the anchor region. Thus we keep pixels that we can use as center of our
# anchor region knowing that it will not overlap with the sample region
# and it will not be outside of bounds
masked_img = cannied_img
# Clip between the bounds
left = sorted(
(0, sample_region[0] * whole_img.shape[0] - (dc_shape[0] / 2),
whole_img.shape[0]))[1]
right = sorted(
(0, sample_region[2] * whole_img.shape[0] + (dc_shape[0] / 2),
whole_img.shape[0]))[1]
top = sorted((0, sample_region[1] * whole_img.shape[1] - (dc_shape[1] / 2),
whole_img.shape[1]))[1]
bottom = sorted(
(0, sample_region[3] * whole_img.shape[1] + (dc_shape[1] / 2),
whole_img.shape[1]))[1]
masked_img[left:right, top:bottom].fill(0)
masked_img[0:(dc_shape[0] / 2), :].fill(0)
masked_img[:, 0:(dc_shape[1] / 2)].fill(0)
masked_img[masked_img.shape[0] -
(dc_shape[0] / 2):masked_img.shape[0], :].fill(0)
masked_img[:, masked_img.shape[1] -
(dc_shape[1] / 2):masked_img.shape[1]].fill(0)
# Find indices of edge pixels
occurrences_indices = numpy.where(masked_img == 255)
X = numpy.matrix(occurrences_indices[0]).T
Y = numpy.matrix(occurrences_indices[1]).T
occurrences = numpy.hstack([X, Y])
# If there is such a pixel outside of the sample region and there is enough
# space according to dc_shape, use the masked image and calculate the anchor
# region roi
if len(occurrences) > 0:
# Enough space outside of the sample region
anchor_roi = ((occurrences[0, 0] -
(dc_shape[0] / 2)) / whole_img.shape[0],
(occurrences[0, 1] -
(dc_shape[1] / 2)) / whole_img.shape[1],
(occurrences[0, 0] +
(dc_shape[0] / 2)) / whole_img.shape[0],
(occurrences[0, 1] +
(dc_shape[1] / 2)) / whole_img.shape[1])
else:
# Not enough space outside of the sample region
# Pick a random pixel
cannied_img = cv2.Canny(uint8_img, 100, 200)
# Find indices of edge pixels
occurrences_indices = numpy.where(cannied_img == 255)
X = numpy.matrix(occurrences_indices[0]).T
Y = numpy.matrix(occurrences_indices[1]).T
occurrences = numpy.hstack([X, Y])
anchor_roi = ((occurrences[0, 0] -
(dc_shape[0] / 2)) / whole_img.shape[0],
(occurrences[0, 1] -
(dc_shape[1] / 2)) / whole_img.shape[1],
(occurrences[0, 0] +
(dc_shape[0] / 2)) / whole_img.shape[0],
(occurrences[0, 1] +
(dc_shape[1] / 2)) / whole_img.shape[1])
return anchor_roi