def closing(image):
"""Numerical closing of the image
Parameters
----------
image: string,
input image
returns
-------
filename: string,
path of closed image
"""
from scipy.ndimage.morphology import grey_closing
# from nilearn.plotting import plot_anat
import nibabel as nib
data = nib.load(image).get_data()
data_ = grey_closing(data, size=(3, 3, 3))
img = nib.Nifti1Image(data_, nib.load(image).affine)
print(np.sum((data - data_)**2))
#cut_coords = [-45, 22, 22]
#plot_anat(image, cut_coords=cut_coords, dim=-1)
#plot_anat(img, cut_coords=cut_coords, dim=-1)
filename = os.path.join(os.path.dirname(image),
os.path.basename(image)[:-4] + '_closed.nii.gz')
img.to_filename(filename)
return filename
def mkoutersurf(image, radius, outfile):
#radius information is currently ignored
#it is a little tougher to deal with the morphology in python
fill = nib.load( image )
filld = fill.get_data()
filld[filld==1] = 255
gaussian = np.ones((2,2))*.25
image_f = np.zeros((256,256,256))
for slice in xrange(256):
temp = filld[:,:,slice]
image_f[:,:,slice] = convolve(temp, gaussian, 'same')
image2 = np.zeros((256,256,256))
image2[np.where(image_f <= 25)] = 0
image2[np.where(image_f > 25)] = 255
strel15 = generate_binary_structure(3, 1)
BW2 = grey_closing(image2, structure=strel15)
thresh = np.max(BW2)/2
BW2[np.where(BW2 <= thresh)] = 0
BW2[np.where(BW2 > thresh)] = 255
v, f = marching_cubes(BW2, 100)
v2 = np.transpose(
np.vstack( ( 128 - v[:,0],
v[:,2] - 128,
128 - v[:,1], )))
write_surface(outfile, v2, f)
def closing(parameters):
"""Calculates morphological closing of a greyscale image.
This is equal to performing a dilation and then an erosion.
It wraps `scipy.ndimage.morphology.grey_closing`. The `footprint`,
`structure`, `output`, `mode`, `cval` and `origin` options are not
supported.
Keep in mind that `mode` and `cval` influence the results. In this case
the default mode is used, `reflect`.
:param parameters['data'][0]: input array
:type parameters['data'][0]: numpy.array
:param parameters['size']: which neighbours to take into account, defaults
to (3, 3) a.k.a. numpy.ones((3, 3))
:type parameters['size']: list
:return: numpy.array
"""
data = parameters['data'][0]
size = tuple(parameters['size'])
return morphology.grey_closing(data, size=size)
def skeletonize(self, image):
image = grey_closing(image, footprint=circle(8), mode='constant', cval=0.0)
image = add_zero_mat(image)
prev_binary_image = np.zeros_like(image)
image_bit_depth = (image.dtype.itemsize * 8) / 2
print "image_bit_depth: " + str(image_bit_depth)
#image_thresholds = range(2**image_bit_depth,-1,-16)
image_thresholds = [2**x for x in range(image_bit_depth, 3, -1)] + range(15, 0, -1)
print "image_thresholds: " + str(image_thresholds)
for curr_threshold in image_thresholds:
print "curr_threshold: " + str(curr_threshold)
curr_thresh_image = threshold(image, curr_threshold)
curr_binary_image = curr_thresh_image.astype(np.bool).astype(np.int)
imsave(skeleton_images_path + "binary_" + str(curr_threshold) + ".png", curr_binary_image)
curr_sum_image = (prev_binary_image + curr_binary_image)
curr_skeleton_image = self.thin_pixels(curr_sum_image)
imsave(skeleton_images_path + "skeleton_" + str(curr_threshold) + ".png", curr_skeleton_image)
print "curr_skeleton max: " + str(curr_skeleton_image.max())
prev_binary_image = curr_skeleton_image
return remove_zero_mat(prev_binary_image)
def processAlgorithm(self, parameters, context, feedback):
"""Here is where the processing itself takes place."""
INPUT_RASTER = self.parameterAsRasterLayer(parameters,
self.INPUT_RASTER, context)
#INPUT_RASTER = self.getParameterValue(self.INPUT_RASTER)
OUTPUT_RASTER = self.parameterAsOutputLayer(parameters,
self.OUTPUT_RASTER,
context)
CLOSING_SIZE = self.parameterAsInt(parameters, self.CLOSING_SIZE,
context)
"""
MEDIAN_ITER = self.parameterAsInt(parameters, self.MEDIAN_ITER, context)
MEDIAN_SIZE = self.parameterAsInt(parameters, self.MEDIAN_SIZE, context)
# First we create the output layer. The output value entered by
# the user is a string containing a filename, so we can use it
# directly
#from scipy import ndimage
#import gdal
"""
INPUT_RASTER_src = INPUT_RASTER.source()
# feedback.pushInfo(str(OUTPUT_RASTER))
#QgsMessageLog.logMessage('output is: '+str(OUTPUT_RASTER))
from scipy.ndimage.morphology import grey_closing
data, im = dataraster.open_data_band(INPUT_RASTER_src)
proj = data.GetProjection()
geo = data.GetGeoTransform()
d = data.RasterCount
total = 100 / (d * 1)
outFile = dataraster.create_empty_tiff(OUTPUT_RASTER, im, d, geo, proj)
iterPos = 0
for i in range(d):
# Read data from the right band
# pbNow+=1
# pb.setValue(pbNow)
tempBand = data.GetRasterBand(i + 1).ReadAsArray()
tempBand = grey_closing(tempBand,
size=(CLOSING_SIZE, CLOSING_SIZE))
#tempBand = tempBand
feedback.setProgress(int(i * total))
# Save bandand outFile
out = outFile.GetRasterBand(i + 1)
out.WriteArray(tempBand)
out.FlushCache()
tempBand = None
return {self.OUTPUT_RASTER: OUTPUT_RASTER}
def Filtro_opening(matrix_imagem):
#gaussian gradiente de magnitude
imagens_filtrada1 = grey_opening(matrix_imagem, size=5)
#imagens_filtrada2=grey_opening(imagens_filtrada1,size=5)
imagens_filtrada = grey_closing(imagens_filtrada1, size=5)
return imagens_filtrada
def get_segmentation_mask(I, mask_color=(1., 1., 1.)):
channel_masks = I.copy()
for c in range(3):
channel_masks[:, :, c] = (I[:, :, c] == mask_color[c]).astype(int)
mask = np.prod(channel_masks, axis=-1)
k = np.ones((3, 3), dtype=np.float32)
mask = spndm.grey_closing(mask, footprint=k)
mask = spndm.grey_opening(mask, footprint=k)
mask = np.clip(mask, 0., 1.)
return mask
def get_segmentation_mask(I, mask_color=(1., 1., 1.)):
from scipy.ndimage import morphology as spndm
channel_masks = I.copy()
for c in range(3):
channel_masks[:, :, c] = (I[:, :, c] == mask_color[c]).astype(int)
mask = np.prod(channel_masks, axis=-1)
k = np.ones((3, 3), dtype=np.float32)
mask = spndm.grey_closing(mask, footprint=k)
mask = spndm.grey_opening(mask, footprint=k)
mask = np.clip(mask, 0., 1.)
return mask
def make_outer_surf(orig_pial, image, radius, outfile):
'''
Make outer surface based on a pial volume and radius,
write to surface in outfile.
Args:
orig_pial: pial surface (e.g. lh.pial)
image: filled lh or rh pial image (e.g. lh.pial.filled.mgz)
radius: radius for smoothing (currently ignored)
outfile: surface file to write data to
Original code from ielu (https://github.com/aestrivex/ielu)
'''
#radius information is currently ignored
#it is a little tougher to deal with the morphology in python
pial_surf = nib.freesurfer.read_geometry(orig_pial, read_metadata=True)
volume_info = pial_surf[2]
fill = nib.load(image)
filld = fill.get_data()
filld[filld == 1] = 255
gaussian = np.ones((2, 2)) * .25
image_f = np.zeros((256, 256, 256))
for slice in xrange(256):
temp = filld[:, :, slice]
image_f[:, :, slice] = convolve(temp, gaussian, 'same')
image2 = np.zeros((256, 256, 256))
image2[np.where(image_f <= 25)] = 0
image2[np.where(image_f > 25)] = 255
strel15 = generate_binary_structure(3, 1)
BW2 = grey_closing(image2, structure=strel15)
thresh = np.max(BW2) / 2
BW2[np.where(BW2 <= thresh)] = 0
BW2[np.where(BW2 > thresh)] = 255
# v, f = marching_cubes(BW2, 100)
v, f = measure.marching_cubes(BW2, 100)
v2 = np.transpose(
np.vstack((
128 - v[:, 0],
v[:, 2] - 128,
128 - v[:, 1],
)))
write_surface(outfile, v2, f, volume_info=volume_info)
def estimate_xheight(line, scale=1.0, debug=0):
"""Estimates the xheight of a line based on image processing and
filtering."""
vgrad = morphology.grey_closing(line, (1, int(scale * 40)))
vgrad = filters.gaussian_filter(vgrad, (2, int(scale * 60)), (1, 0))
if amin(vgrad) > 0 or amax(vgrad) < 0: raise Exception("bad line")
if debug: imshow(vgrad)
proj = sum(vgrad, 1)
proj = filters.gaussian_filter(proj, 0.5)
top = argmax(proj)
bottom = argmin(proj)
return bottom - top, bottom
def estimate_xheight(line,scale=1.0,debug=0):
"""Estimates the xheight of a line based on image processing and
filtering."""
vgrad = morphology.grey_closing(line,(1,int(scale*40)))
vgrad = filters.gaussian_filter(vgrad,(2,int(scale*60)),(1,0))
if amin(vgrad)>0 or amax(vgrad)<0: raise BadImage("bad line")
if debug: imshow(vgrad)
proj = sum(vgrad,1)
proj = filters.gaussian_filter(proj,0.5)
top = argmax(proj)
bottom = argmin(proj)
return bottom-top,bottom
def estimate_baseline(line, order=3):
"""Compute the baseline by fitting a polynomial to the gradient.
TODO: use robust fitting, special case very short line, limit parameter ranges"""
line = line * 1.0 / amax(line)
vgrad = morphology.grey_closing(line, (1, 40))
vgrad = filters.gaussian_filter(vgrad, (2, 60), (1, 0))
if amin(vgrad) > 0 or amax(vgrad) < 0: raise BadLine()
h, w = vgrad.shape
ys = argmin(vgrad, axis=0)
xs = arange(w)
baseline = polyfit(xs, ys, order)
print baseline
return baseline
def estimate_baseline(line,order=3):
"""Compute the baseline by fitting a polynomial to the gradient.
TODO: use robust fitting, special case very short line, limit parameter ranges"""
line = line*1.0/amax(line)
vgrad = morphology.grey_closing(line,(1,40))
vgrad = filters.gaussian_filter(vgrad,(2,60),(1,0))
if amin(vgrad)>0 or amax(vgrad)<0: raise BadImage()
h,w = vgrad.shape
ys = argmin(vgrad,axis=0)
xs = arange(w)
baseline = polyfit(xs,ys,order)
print baseline
return baseline
def latin_mask(line, scale=1.0, r=1.2, debug=0):
"""Estimate a mask that covers letters and diacritics of a text
line for Latin alphabets."""
vgrad = morphology.grey_closing(1.0 * line, (1, int(scale * 40)))
vgrad = filters.gaussian_filter(vgrad, (2, int(scale * 60)), (1, 0))
tops = argmax(vgrad, 0)
bottoms = argmin(vgrad, 0)
mask = zeros(line.shape)
xheight = mean(bottoms - tops)
for i in range(len(bottoms)):
d = bottoms[i] - tops[i]
y0 = int(maximum(0, bottoms[i] - r * d))
mask[y0:bottoms[i], i] = 1
return mask
def latin_mask(line,scale=1.0,r=1.2,debug=0):
"""Estimate a mask that covers letters and diacritics of a text
line for Latin alphabets."""
vgrad = morphology.grey_closing(1.0*line,(1,int(scale*40)))
vgrad = filters.gaussian_filter(vgrad,(2,int(scale*60)),(1,0))
tops = argmax(vgrad,0)
bottoms = argmin(vgrad,0)
mask = zeros(line.shape)
xheight = mean(bottoms-tops)
for i in range(len(bottoms)):
d = bottoms[i]-tops[i]
y0 = int(maximum(0,bottoms[i]-r*d))
mask[y0:bottoms[i],i] = 1
return mask
def extract_watermark(audio_file, interactive=False):
"""
Extracts the watermark from the spectrogram of the given audio file
:param audio_file: path to wav file
:param interactive: activates plotting
:return: watermark as text, or None if the watermark could not be extracted
"""
# Convert audio file to wav if necessary
wavFile = convert_to_wav(audio_file)
fs, data = wavfile.read(wavFile)
data = data.astype(np.float) / np.max(np.abs(data))
window_length = 1024
nfft = window_length
h = window_length // 4
spectrogram, f, t = stft(data, window_length, h, nfft, fs)
if interactive:
plot_spectrogram(spectrogram)
# Convert to PIL image in order to use optical character recognition
# Flip upside down due to the usual way in which we view a spectrogram
ocr_image = np.flipud(np.abs(spectrogram))
# Do some image enhancement
ocr_image[ocr_image < 0.2] = 0
ocr_image = grey_closing(ocr_image, (5, 2))
ocr_image = grey_erosion(ocr_image, (3, 5))
# Convert to 8 bit image
ocr_image = np.uint8(ocr_image / np.max(ocr_image) * 255 * 10)[20:120, :]
ocr_image[ocr_image > 5] = 255
# Enlarge image by interpolation
# ocr_image = imresize(ocr_image, (ocr_image.shape[0] * 8, ocr_image.shape[1] * 8), interp="bilinear")
if interactive:
# Show for debugging purposes
plt.imshow(ocr_image)
plt.show()
ocr_image = Image.fromarray(ocr_image)
ocr_image_filename = "test.png"
ocr_image.save(ocr_image_filename, format="png")
# watermark = ocr.tesseract(ocr_image)
watermark = ocr_space(ocr_image_filename)
# ocr_image.save("test.png", format="png")
return watermark
def refined_seeding(a, maximum_height=0, grey_close_radius=1,
binary_open_radius=1, binary_close_radius=1, minimum_size=0):
"""Perform morphological operations to get good segmentation seeds."""
if grey_close_radius > 0:
strel = diamond_se(grey_close_radius, a.ndim)
a = grey_closing(a, footprint=strel)
s = (a <= maximum_height)
if binary_open_radius > 0:
strel = diamond_se(binary_open_radius, s.ndim)
s = binary_opening(s, structure=strel)
if binary_close_radius > 0:
strel = diamond_se(binary_close_radius, s.ndim)
s = binary_closing(s, structure=strel)
s = remove_small_connected_components(s, minimum_size)
return label(s)[0]
def morph_filter(raw_image, closing_radius, opening_radius):
# Create circular masks
close_footprint, open_footprint = [[[0 for ii in range(r * 2 + 1)] for i in range(r * 2 + 1)] for r in [closing_radius, opening_radius]]
for fp in [close_footprint, open_footprint]:
r = (len(fp) - 1) / 2
for i in range(len(fp)):
for ii in range(len(fp)):
if (i - r) ** 2 + (ii - r) ** 2 <= r ** 2: fp[i][ii] = 1
# Perform filtering
filtered_image = raw_image
filtered_image = grey_closing(filtered_image, footprint=close_footprint)
filtered_image = grey_opening(filtered_image, footprint=open_footprint)
return filtered_image
def depthmap_flow_nav(d_im):
# calculates a crude depth flow field and identifies a likely clear path
# by template matching to a gaussian distance function
global cX, cY
global yaw_error
global est_dist
# create nxn zeros and appropriate kernel
kernlen = 321
dirac_im = np.zeros((kernlen, kernlen))
# set element at the middle to one, a dirac delta
dirac_im[kernlen//2, kernlen//2] = 1
# gaussian-smooth the dirac, resulting in a gaussian filter:
gauss_template = cv2.GaussianBlur(dirac_im, (kernlen, kernlen), 0)
# normalise
max_g = max(gauss_template.max(axis=1))
gauss_display = np.array(255*gauss_template/max_g, dtype=np.uint8)
# filter the distance output to remove discontinuities and approximate a flow field
d_im_filt = scmorph.grey_closing(d_im, size=(7, 7))
blur = cv2.GaussianBlur(d_im_filt, (71, 71), 0)
# we may want to restrict our analysis to a central 'band' of the image
# can use a mask in the template match for this
blur = np.array(blur, dtype=np.uint8)
# Cross correlate a gaussian peaked function of size < image:
template_match = cv2.matchTemplate(blur, gauss_display, cv2.TM_CCORR_NORMED)
template_match = cv2.normalize(template_match, 0, 1, cv2.NORM_MINMAX)
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(template_match)
# max_loc gives top left of template match
cX = max_loc[0] + kernlen // 2
cY = max_loc[1] + kernlen // 2
# distance: examine distance at max_loc to estimate whether we have hit a wall? This doesn't work very well!
vis_cent = blur[(cX-8):(cX+8), (cY-8):(cY+8)]
vis_cent = vis_cent.astype('float64')
vis_cent[vis_cent < 5 ] = np.nan
est_dist = np.nanmean(vis_cent)
yaw_error = cX - 640/2 # change this for different depth image size
def refined_seeding(a,
maximum_height=0,
grey_close_radius=1,
binary_open_radius=1,
binary_close_radius=1,
minimum_size=0):
"""Perform morphological operations to get good segmentation seeds."""
if grey_close_radius > 0:
strel = diamond_se(grey_close_radius, a.ndim)
a = grey_closing(a, footprint=strel)
s = (a <= maximum_height)
if binary_open_radius > 0:
strel = diamond_se(binary_open_radius, s.ndim)
s = binary_opening(s, structure=strel)
if binary_close_radius > 0:
strel = diamond_se(binary_close_radius, s.ndim)
s = binary_closing(s, structure=strel)
s = remove_small_connected_components(s, minimum_size)
return label(s)[0]
def gen(py_dev):
running = True
frameint = 5
framecount = 0
global yaw_error
while running:
framecount += 1
py_dev.wait_for_frame()
c_im = py_dev.colour
rgb_im = c_im[..., ::-1]
# try to scale with minimal wrap over effective range - fairly heuristic
# Note that most colormaps give darker values to closer items
# we may want to invert this, intuitively
d_im = py_dev.depth * 0.05
# close holes without removing segmentation by doing it before converting to image
d_im_filt = scmorph.grey_closing(d_im, size=(7, 7))
d_im_col = cv2.applyColorMap(d_im.astype(np.uint8), cv2.COLORMAP_HOT)
# every nth frame, update direction by analysing depth map
# two options: segmentation or gradient. Segmentation is a problem with very noisy images
if framecount > frameint:
#thread.start_new_thread(depthmap_seg_nav, (d_im_col, ))
thread.start_new_thread(depthmap_flow_nav, (d_im_filt, ))
framecount = 1
# eventually replace this with an arrow indicating desired direction
cv2.circle(d_im_col, (cX, cY), 7, (255, 255, 255), -1)
cv2.putText(d_im_col, str(yaw_error), (cX - 20, cY - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 2)
#cd = np.concatenate((blur, gauss_template), axis=1)
cd = np.concatenate((rgb_im, d_im_col), axis=1)
ret, frame = cv2.imencode('.jpg', cd)
# this is pretty slow over wifi
jpeg_encode = frame.tobytes()
yield (b'--frame\r\n'
b'Content-Type: image/jpeg\r\n\r\n' + jpeg_encode + b'\r\n\r\n')
def mkoutersurf(image, radius, outfile):
#radius information is currently ignored
#it is a little tougher to deal with the morphology in python
fill = nib.load(image)
filld = fill.get_data()
filld[filld == 1] = 255
gaussian = np.ones((2, 2)) * .25
image_f = np.zeros((256, 256, 256))
for slice in range(256):
temp = filld[:, :, slice]
image_f[:, :, slice] = convolve(temp, gaussian, 'same')
image2 = np.zeros((256, 256, 256))
image2[np.where(image_f <= 25)] = 0
image2[np.where(image_f > 25)] = 255
strel15 = generate_binary_structure(3, 1)
BW2 = grey_closing(image2, structure=strel15)
thresh = np.max(BW2) / 2
BW2[np.where(BW2 <= thresh)] = 0
BW2[np.where(BW2 > thresh)] = 255
v, f, _, _ = measure.marching_cubes_lewiner(BW2, 100)
v2 = np.transpose(
np.vstack((
128 - v[:, 0],
v[:, 2] - 128,
128 - v[:, 1],
)))
write_surface(outfile, v2, f)
def make_outer_surf(
orig_pial: Union[str, Path],
image: Union[str, Path],
output_fpath: Union[str, Path],
outer_surface_sphere: float = 15,
):
"""Create outer surface of a pial volume.
Make outer surface based on a pial volume and radius,
write to surface in outfile.
Parameters
----------
orig_pial : str | pathlib.Path
Pial surface (e.g. lh.pial)
image : str | pathlib.Path
Filled lh or rh pial image (e.g. lh.pial.filled.mgz)
output_fpath : str | pathlib.Path
surface file to write data to
outer_surface_sphere : float | None
radius for smoothing in mm (default=15). diameter of the sphere used
by make_outer_surface to close the sulci using morphological operations.
Ignored currently. Corresponds to ``se=strel('sphere',se_diameter);``
in Matlab. See [1].
References
----------
.. [1] See FieldTrip Toolbox ``make_outer_surface`` function inside
``prepare_mesh_cortexhull.m`` file.
.. [2] https://github.com/aestrivex/ielu
"""
from scipy.signal import convolve
from scipy.ndimage.morphology import grey_closing, generate_binary_structure
from mne import write_surface
from mcubes import marching_cubes
# radius information is currently ignored
# it is a little tougher to deal with the morphology in python
# load original pial surface to get the volume information
pial_surf = nb.freesurfer.read_geometry(orig_pial, read_metadata=True)
volume_info = pial_surf[2]
# load filled pial image
fill = nb.load(image)
filld = fill.get_data()
filld[filld == 1] = 255
# apply a very soft Gaussian filter with sigma = 1mm to
# facilitate the closing
gaussian = np.ones((2, 2)) * 0.25
# initialize image cube array
image_f = np.zeros((256, 256, 256))
# initialize a thresholded image
image2 = np.zeros((256, 256, 256))
# for each slice, convolve the Gaussian filter on the
# filled image
for slice in range(256):
temp = filld[:, :, slice]
image_f[:, :, slice] = convolve(temp, gaussian, "same")
# thresholded image based on value of 25
image2[np.where(image_f <= 25)] = 0
image2[np.where(image_f > 25)] = 255
strel15 = generate_binary_structure(3, 1)
# run multi-dimensional grayscale closing of the image
BW2 = grey_closing(image2, structure=strel15)
thresh = np.max(BW2) / 2
BW2[np.where(BW2 <= thresh)] = 0
BW2[np.where(BW2 > thresh)] = 255
# apply marching cubes algorithm to get
# vertices and faces
v, f = marching_cubes(BW2, 100)
# in order to cope with the different orientation
v2 = np.transpose(
np.vstack((
128 - v[:, 0],
v[:, 2] - 128,
128 - v[:, 1],
)))
write_surface(output_fpath, v2, f, volume_info=volume_info)
def build_iscat_training(bf_filepaths, iscat_filepaths, sampling=4):
"""Creates iscat training data and target in data/iscat_seg/[REF_FRAMES / MASKS] for the iSCAT cell segmentation task
ARGS:
bf_filepaths (list(str)): filepaths of all the bright field images to input as returned by utilitiues.load_data_paths()
iscat_filepaths (list(str)): filepaths of all the iscat images to input as returned by utilitiues.load_data_paths()
sampling (int): sampling interval of the saved images (lower storage footprint)
"""
OUT_PATH = DATA_PATH + 'iscat_seg/'
os.makedirs(os.path.join(OUT_PATH, 'REF_FRAMES/'), exist_ok=True)
os.makedirs(os.path.join(OUT_PATH, 'MASKS/'), exist_ok=True)
# Range of non filtered elements [px]
min_size, max_size = 1, 13
iscat_stacks = (utilities.load_imgs(path) for path in iscat_filepaths)
bf_stacks = (utilities.load_imgs(path) for path in bf_filepaths)
# Returns the metadata of the exwperiments such as frame rate
metadatas = get_experiments_metadata(iscat_filepaths)
if torch.cuda.is_available():
device = torch.cuda.current_device()
torch.cuda.set_device(device)
print("Running on: {:s}".format(torch.cuda.get_device_name(device)))
cuda = torch.device('cuda')
else:
# Doesn't run on CPU only machines comment if no GPU
print("No CUDA device found")
sys.exit(1)
unet = UNetCell(1, 1, device=cuda, bilinear_upsampling=False)
unet.load_state_dict(torch.load('outputs/saved_models/bf_unet.pth'))
for i, (bf_stack, iscat_stack,
metadata) in enumerate(zip(bf_stacks, iscat_stacks, metadatas)):
if i < 45: continue
bf_stack = bf_stack.astype('float32')
print(bf_stack.shape)
if bf_stack.shape[1:] != iscat_stack.shape[1:]:
bf_stack = processing.coregister(bf_stack, 1.38)
print(bf_stack.shape)
normalize(bf_stack)
# Samples iscat image to correct for the difference in framefate
iscat_stack = iscat_stack[::sampling * int(metadata['iscat_fps'] /
metadata['tirf_fps'])]
torch_stack = torch.from_numpy(bf_stack).unsqueeze(1).cuda()
mask = unet.predict_stack(
torch_stack).detach().squeeze().cpu().numpy() > 0.05
mask = morphology.grey_erosion(mask * 255,
structure=processing.structural_element(
'circle', (3, 5, 5)))
mask = morphology.grey_closing(mask,
structure=processing.structural_element(
'circle', (3, 7, 7)))
mask = (mask > 50).astype('uint8')
# Median filtering and normalization
iscat_stack = processing.image_correction(iscat_stack)
# Contrast enhancement
iscat_stack = processing.enhance_contrast(iscat_stack,
'stretching',
percentile=(1, 99))
# Fourier filtering of image
iscat_stack = processing.fft_filtering(iscat_stack, min_size, max_size,
True)
iscat_stack = processing.enhance_contrast(iscat_stack,
'stretching',
percentile=(3, 97))
for j in range(0, min(iscat_stack.shape[0], mask.shape[0]), sampling):
if iscat_stack[j].shape == mask[j].shape:
# Doesn't save images without detected cells
if mask[j].max() == 0: continue
print("\rSaving to stack_{}_{}.png".format(i + 1, j + 1),
end=' ' * 5)
tifffile.imsave(
os.path.join(OUT_PATH, 'REF_FRAMES/',
"stack_{}_{}.png".format(i + 1, j + 1)),
rescale(iscat_stack[j]))
tifffile.imsave(
os.path.join(OUT_PATH, 'MASKS/',
"mask_{}_{}.png".format(i + 1, j + 1)),
mask[j] * 255)
else:
print("Error, shape: {}, {}".format(iscat_stack[j].shape,
mask[j].shape))
break
print('')
def gray_closing(self, *args, **kw):
'''see scipy.ndimage.morphology.grey_closing'''
return Image(_morphology.grey_closing(self, *args, **kw)).convert_type(self.dtype)
def mark_orders(
im,
min_cluster=None,
min_width=None,
filter_size=None,
noise=None,
opower=4,
border_width=None,
degree_before_merge=2,
regularization=0,
closing_shape=(5, 5),
opening_shape=(2, 2),
plot=False,
plot_title=None,
manual=True,
auto_merge_threshold=0.9,
merge_min_threshold=0.1,
sigma=0,
):
"""Identify and trace orders
Parameters
----------
im : array[nrow, ncol]
order definition image
min_cluster : int, optional
minimum cluster size in pixels (default: 500)
filter_size : int, optional
size of the running filter (default: 120)
noise : float, optional
noise to filter out (default: 8)
opower : int, optional
polynomial degree of the order fit (default: 4)
border_width : int, optional
number of pixels at the bottom and top borders of the image to ignore for order tracing (default: 5)
plot : bool, optional
wether to plot the final order fits (default: False)
manual : bool, optional
wether to manually select clusters to merge (strongly recommended) (default: True)
Returns
-------
orders : array[nord, opower+1]
order tracing coefficients (in numpy order, i.e. largest exponent first)
"""
# Convert to signed integer, to avoid underflow problems
im = np.asanyarray(im)
im = im.astype(int)
if filter_size is None:
col = im[:, im.shape[0] // 2]
col = median_filter(col, 5)
threshold = np.percentile(col, 90)
npeaks = find_peaks(col, height=threshold)[0].size
filter_size = im.shape[0] // (npeaks * 2)
logger.info("Median filter size, estimated: %i", filter_size)
elif filter_size <= 0:
raise ValueError(f"Expected filter size > 0, but got {filter_size}")
if border_width is None:
# find width of orders, based on central column
col = im[:, im.shape[0] // 2]
col = median_filter(col, 5)
idx = np.argmax(col)
width = peak_widths(col, [idx])[0][0]
border_width = int(np.ceil(width))
logger.info("Image border width, estimated: %i", border_width)
elif border_width < 0:
raise ValueError(f"Expected border width > 0, but got {border_width}")
if min_cluster is None:
min_cluster = im.shape[1] // 4
logger.info("Minimum cluster size, estimated: %i", min_cluster)
elif not np.isscalar(min_cluster):
raise TypeError(
f"Expected scalar minimum cluster size, but got {min_cluster}")
if min_width is None:
min_width = 0.25
if min_width == 0:
pass
elif isinstance(min_width, (float, np.floating)):
min_width = int(min_width * im.shape[0])
logger.info("Minimum order width, estimated: %i", min_width)
# im[im < 0] = np.ma.masked
blurred = np.ma.filled(im, fill_value=0)
blurred = grey_closing(blurred, 5)
# blur image along columns, and use the median + blurred + noise as threshold
blurred = gaussian_filter1d(blurred, filter_size, axis=0)
if noise is None:
tmp = np.abs(blurred.flatten())
noise = np.percentile(tmp, 5)
logger.info("Background noise, estimated: %f", noise)
elif not np.isscalar(noise):
raise TypeError(f"Expected scalar noise level, but got {noise}")
mask = im > blurred + noise
# remove borders
if border_width != 0:
mask[:border_width, :] = mask[-border_width:, :] = False
mask[:, :border_width] = mask[:, -border_width:] = False
# remove masked areas with no clusters
mask = np.ma.filled(mask, fill_value=False)
# close gaps inbetween clusters
struct = np.full(closing_shape, 1)
mask = morphology.binary_closing(mask, struct, border_value=1)
# remove small lonely clusters
struct = np.full(opening_shape, 1)
# struct = morphology.generate_binary_structure(2, 1)
mask = morphology.binary_opening(mask, struct)
# label clusters
clusters, _ = label(mask)
# remove small clusters
sizes = np.bincount(clusters.ravel())
mask_sizes = sizes > min_cluster
mask_sizes[
0] = True # This is the background, which we don't need to remove
for i in np.arange(len(sizes))[~mask_sizes]:
clusters[clusters == i] = 0
# # Reorganize x, y, clusters into a more convenient "pythonic" format
# # x, y become dictionaries, with an entry for each order
# # n is just a list of all orders (ignore cluster == 0)
n = np.unique(clusters)
n = n[n != 0]
x = {i: np.where(clusters == c)[0] for i, c in enumerate(n)}
y = {i: np.where(clusters == c)[1] for i, c in enumerate(n)}
def best_fit_degree(x, y):
L1 = np.sum((np.polyval(np.polyfit(y, x, 1), y) - x)**2)
L2 = np.sum((np.polyval(np.polyfit(y, x, 2), y) - x)**2)
# aic1 = 2 + 2 * np.log(L1) + 4 / (x.size - 2)
# aic2 = 4 + 2 * np.log(L2) + 12 / (x.size - 3)
if L1 < L2:
return 1
else:
return 2
if sigma > 0:
degree = {i: best_fit_degree(x[i], y[i]) for i in x.keys()}
bias = {i: np.polyfit(y[i], x[i], deg=degree[i])[-1] for i in x.keys()}
n = list(x.keys())
yt = np.concatenate([y[i] for i in n])
xt = np.concatenate([x[i] - bias[i] for i in n])
coef = np.polyfit(yt, xt, deg=degree_before_merge)
res = np.polyval(coef, yt)
cutoff = sigma * (res - xt).std()
# DEBUG plot
# uy = np.unique(yt)
# mask = np.abs(res - xt) > cutoff
# plt.plot(yt, xt, ".")
# plt.plot(yt[mask], xt[mask], "r.")
# plt.plot(uy, np.polyval(coef, uy))
# plt.show()
#
m = {
i: np.abs(np.polyval(coef, y[i]) - (x[i] - bias[i])) < cutoff
for i in x.keys()
}
k = max(x.keys()) + 1
for i in range(1, k):
new_img = np.zeros(im.shape, dtype=int)
new_img[x[i][~m[i]], y[i][~m[i]]] = 1
clusters, _ = label(new_img)
x[i] = x[i][m[i]]
y[i] = y[i][m[i]]
if len(x[i]) == 0:
del x[i], y[i]
nnew = np.max(clusters)
if nnew != 0:
xidx, yidx = np.indices(im.shape)
for j in range(1, nnew + 1):
xn = xidx[clusters == j]
yn = yidx[clusters == j]
if xn.size >= min_cluster:
x[k] = xn
y[k] = yn
k += 1
# plt.imshow(clusters, origin="lower")
# plt.show()
if plot: # pragma: no cover
title = "Identified clusters"
if plot_title is not None:
title = f"{plot_title}\n{title}"
plt.title(title)
plt.xlabel("x [pixel]")
plt.ylabel("y [pixel]")
clusters = np.ma.zeros(im.shape, dtype=int)
for i in x.keys():
clusters[x[i], y[i]] = i + 1
clusters[clusters == 0] = np.ma.masked
plt.imshow(clusters, origin="lower", cmap="prism")
plt.show()
# Merge clusters, if there are even any possible mergers left
x, y, n = merge_clusters(
im,
x,
y,
n,
manual=manual,
deg=degree_before_merge,
auto_merge_threshold=auto_merge_threshold,
merge_min_threshold=merge_min_threshold,
plot_title=plot_title,
)
if min_width > 0:
sizes = {k: v.max() - v.min() for k, v in y.items()}
mask_sizes = {k: v > min_width for k, v in sizes.items()}
for k, v in mask_sizes.items():
if not v:
del x[k]
del y[k]
n = x.keys()
orders = fit_polynomials_to_clusters(x, y, n, opower)
# sort orders from bottom to top, using relative position
def compare(i, j):
_, xi, i_left, i_right = i
_, xj, j_left, j_right = j
if i_right < j_left or j_right < i_left:
return xi.mean() - xj.mean()
left = max(i_left, j_left)
right = min(i_right, j_right)
return xi[left:right].mean() - xj[left:right].mean()
xp = np.arange(im.shape[1])
keys = [(c, np.polyval(orders[c], xp), y[c].min(), y[c].max())
for c in x.keys()]
keys = sorted(keys, key=cmp_to_key(compare))
key = [k[0] for k in keys]
n = np.arange(len(n), dtype=int)
x = {c: x[key[c]] for c in n}
y = {c: y[key[c]] for c in n}
orders = np.array([orders[key[c]] for c in n])
column_range = np.array([[np.min(y[i]), np.max(y[i]) + 1] for i in n])
if plot: # pragma: no cover
plot_orders(im, x, y, n, orders, column_range, title=plot_title)
return orders, column_range
def main():
# parse cmd arguments
parser = getParser()
parser.parse_args()
args = getArguments(parser)
# prepare logger
logger = Logger.getInstance()
if args.debug: logger.setLevel(logging.DEBUG)
elif args.verbose: logger.setLevel(logging.INFO)
logger.info('Executing weighted viscous morphology with {} ({} bins).'.format(','.join(map(str, args.func)), len(args.func)))
# iterate over input images
for image in args.images:
# build output file name
image_viscous_name = args.folder + '/' + image.split('/')[-1][:-4] + '_wviscous_' + '_'.join(map(str, args.func))
image_viscous_name += image.split('/')[-1][-4:]
# check if output file exists
if not args.force:
if os.path.exists(image_viscous_name):
logger.warning('The output file {} already exists. Skipping this image.'.format(image_viscous_name))
continue
# get and prepare image data
logger.info('Loading image {} using NiBabel...'.format(image))
image_gradient = load(image)
# get and prepare image data
image_gradient_data = scipy.squeeze(image_gradient.get_data())
# prepare result image and extract required attributes of input image
if args.debug:
logger.debug('Intensity range of gradient image is ({}, {})'.format(image_gradient_data.min(), image_gradient_data.max()))
# create gradient images flattened histogram
bins = hist_flatened(image_gradient_data, len(args.func))
logger.debug('{} bins created'.format(len(bins) -1))
# check if the number of bins is consistent
if len(args.func) != len(bins) - 1:
raise Exception('Inconsistency between the number of requested and created bins ({} to {})'.format(args.sections, len(bins) - 1))
# prepare result file
image_viscous_data = image_gradient_data
# transform the gradient images topography
logger.info('Applying the viscous morphological operations on {} sections...'.format(len(args.func)))
for sl in range(1, len(args.func) + 1):
# create sphere to use in this step
if 0 >= args.func[sl - 1]: continue # sphere of sizes 0 or below lead to no changes and are not executed
sphere = iterate_structure(generate_binary_structure(3, 1), args.func[sl - 1]).astype(scipy.int_)
# create masks to extract the affected voxels (i.e. the current slice of the topographic image representation)
mask_greater = (image_gradient_data >= bins[sl]) # all voxels with are over the current slice
mask_lower = (image_gradient_data < bins[sl - 1]) # all voxels which are under the current slice
mask_equal = scipy.invert(mask_greater | mask_lower) # all voxels in the current slice
# extract slice
image_threshold_data = image_gradient_data.copy()
image_threshold_data[mask_lower] = 0 # set all voxels under the current slice to zero
image_threshold_data[mask_greater] = image_threshold_data[mask_equal].max() # set all voxels over the current slice to the max of all voxels in the current slice
logger.debug('{} of {} voxels belong to this level.'.format(len(mask_equal.nonzero()[0]), scipy.prod(image_threshold_data.shape)))
# apply the closing with the appropriate sphere
logger.debug('Applying a disk of {} to all values >= {} and < {} (sec {})...'.format(args.func[sl - 1], bins[sl - 1], bins[sl], sl))
image_closed_data = grey_closing(image_threshold_data, footprint=sphere)
# add result of this slice to the general results
image_viscous_data = scipy.maximum(image_viscous_data, image_closed_data)
# save resulting gradient image
logger.info('Saving resulting gradient image as {}...'.format(image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
logger.info('Successfully terminated.')
def main():
# parse cmd arguments
parser = getParser()
parser.parse_args()
args = getArguments(parser)
# prepare logger
logger = Logger.getInstance()
if args.debug: logger.setLevel(logging.DEBUG)
elif args.verbose: logger.setLevel(logging.INFO)
logger.info('Selected viscous type is {}'.format(args.type))
# iterate over input images
for image in args.images:
# get and prepare image data
logger.info('Loading image {} using NiBabel...'.format(image))
image_gradient = load(image)
# get and prepare image data
image_gradient_data = scipy.squeeze(image_gradient.get_data())
logger.debug('Intensity range of gradient image is ({}, {})'.format(
image_gradient_data.min(), image_gradient_data.max()))
# build output file name and check for its existence, if not in sections mode
if 'sections' != args.type:
# build output file name
image_viscous_name = args.folder + '/' + image.split(
'/')[-1][:-4] + '_viscous_{}_sec_{}_ds_{}'.format(
args.type, args.sections, args.dsize)
image_viscous_name += image.split('/')[-1][-4:]
# check if output file exists
if not args.force:
if os.path.exists(image_viscous_name):
logger.warning(
'The output file {} already exists. Skipping this image.'
.format(image_viscous_name))
continue
# execute plain closing i.e. a closing operation over the whole image, if in plain mode
if 'plain' == args.type:
# prepare the disc structure (a ball with a diameter of (args.dsize * 2 + 1))
disc = iterate_structure(generate_binary_structure(3, 1),
args.dsize).astype(scipy.int_)
# apply closing
logger.info('Applying the morphology over whole image at once...')
image_viscous_data = grey_closing(image_gradient_data,
footprint=disc)
# save resulting gradient image
logger.info('Saving resulting gradient image as {}...'.format(
image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
# skip other morphologies
continue
# create gradient images flattened histogram
bins = hist_flatened(image_gradient_data, args.sections)
logger.debug('{} bins created'.format(len(bins) - 1))
# check if the number of bins is consistent
if args.sections != len(bins) - 1:
raise Exception(
'Inconsistency between the number of requested and created bins ({} to {})'
.format(args.sections,
len(bins) - 1))
# prepare result file
image_viscous_data = image_gradient_data
# transform the gradient images topography (Note: the content of one bin is: bins[slice - 1] <= content < bins[slice]
logger.info(
'Applying the viscous morphological operations {} times...'.format(
args.sections))
for slice in range(1, args.sections + 1):
# build output file name and check for its existence, if in sections mode
if 'sections' == args.type:
# build output file name
image_viscous_name = args.folder + '/' + image.split(
'/')[-1][:-4] + '_viscous_{}_sec_{}_ds_{}_sl_{}'.format(
args.type, args.sections, args.dsize, slice)
image_viscous_name += image.split('/')[-1][-4:]
# check if output file exists
if not args.force:
if os.path.exists(image_viscous_name):
logger.warning(
'The output file {} already exists. Skipping this slice.'
.format(image_viscous_name))
continue
# prepare result file
image_viscous_data = image_gradient_data
# create masks to extract the affected voxels (i.e. the current slice of the topographic image representation)
mask_greater = (image_gradient_data >= bins[slice]
) # all voxels with are over the current slice
mask_lower = (image_gradient_data < bins[slice - 1]
) # all voxels which are under the current slice
mask_equal = scipy.invert(
mask_greater | mask_lower) # all voxels in the current slice
if 'mercury' == args.type:
dsize = int((args.dsize / float(args.sections)) * (slice))
disc = iterate_structure(generate_binary_structure(3, 1),
dsize).astype(scipy.int_)
mask_equal_or_greater = mask_equal | mask_greater
image_threshold_data = image_gradient_data * mask_equal_or_greater
elif 'oil' == args.type:
dsize = int((args.dsize / float(args.sections)) *
(args.sections - slice + 1))
disc = iterate_structure(generate_binary_structure(3, 1),
dsize).astype(scipy.int_)
image_threshold_data = image_gradient_data.copy()
mask_equal_or_lower = mask_equal | mask_lower
# set all voxels over the current slice to the max of all voxels in the current slice
image_threshold_data[mask_greater] = image_threshold_data[
mask_equal_or_lower].max()
elif 'sections' == args.type:
dsize = args.dsize
disc = iterate_structure(generate_binary_structure(3, 1),
args.dsize).astype(scipy.int_)
image_threshold_data = image_gradient_data.copy()
# set all voxels under the current slice to zero
image_threshold_data[mask_lower] = 0
# set all voxels over the current slice to the max of all voxels in the current slice
image_threshold_data[mask_greater] = image_threshold_data[
mask_equal].max()
logger.debug('{} of {} voxels belong to this level.'.format(
len(mask_equal.nonzero()[0]),
scipy.prod(image_threshold_data.shape)))
# apply the closing with the appropriate disc size
logger.debug(
'Applying a disk of {} to all values >= {} and < {}...'.format(
dsize, bins[slice - 1], bins[slice]))
image_closed_data = grey_closing(image_threshold_data,
footprint=disc)
# add result of this slice to the general results
image_viscous_data = scipy.maximum(image_viscous_data,
image_closed_data)
# save created output file, if in sections mode
if 'sections' == args.type:
# save resulting gradient image
logger.info('Saving resulting gradient image as {}...'.format(
image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
# save created output file, if not in sections mode
if 'sections' != args.type:
# save resulting gradient image
logger.info('Saving resulting gradient image as {}...'.format(
image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
logger.info('Successfully terminated.')
def main(*args):
if(len(args) < 3):
return 1
git_hash = check_output(['git','rev-parse','HEAD']).strip()
for i in range(2,len(args),2):
nlevels = 256
seed = random.randint(0, sys.maxint)
random.seed(seed)
np.random.seed(seed)
r = np.array(range(0,nlevels))
source = pickle.load(open(args[1],'rb'))
gt_path = args[i]
out_path = args[i+1]
seed_path = os.path.splitext(out_path)[0]+'.seed'
ground = np.genfromtxt(gt_path,dtype='int16')
ground_edges = edges(ground)
with open(seed_path,'wb') as f:
f.write(str(seed)+' '+git_hash)
ncircles = random.randint(0,5)
print('Number of circles: ' + str(ncircles))
nlines = random.randint(0,2)
print('Number of lines: ' + str(nlines))
nscratches = random.randint(ground.min(),ground.max())
print('Number of scratches: ' + str(nscratches))
out = np.zeros(ground.shape,dtype='int16')
dt = distance_transform_edt(np.logical_not(ground_edges))
# supress edges
for p,e in edge_list(ground):
# lower threshold on edge "length"
if e.sum() < 50:
continue
s = binary_dilation(e,iterations=3)
dt[s] += random.choice([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.25, 0.5, 1])
# for c in [ (random.randint(0,ground.shape[0]),
# random.randint(0,ground.shape[1]),
# random.randint(25,75))
# for k in range(0,ncircles) ]:
# dt[matsci.adj.circle(c,out.shape)] += 1
dt = dt.astype('int16')
out[dt==0] = normsamp(source['dist_raw'][0], len(out[dt==0]), 255,128)
out = grey_closing(out, size=(3,3))
for d in source['dist']:
if not d==0:
out[dt==d] = histsamp(source['dist'][d], len(out[dt==d]))
# fill in the rest of the pixels with value from last sampling
m = max(source['dist'].keys())
out[dt>=m-1] = histsamp(source['dist'][m], len(out[dt>=m-1]))
# create varying intensities within grains
for k in range(0,ground.max()):
out[ground==k] += histsamp(source['grain_all'],1)[0]
for l in [ binary_dilation(
line( (random.randint(0,ground.shape[0]),
random.randint(0,ground.shape[1]))
, random.random() * math.pi * 2 # angle
, int(math.sqrt(math.pow(ground.shape[0],2)+
math.pow(ground.shape[1],2))) + 1 # length
, ground.shape))
for k in range(0,nlines) ]:
out[l] += np.clip(map(int,map(round,norm.rvs(loc=64,
scale=32,
size=len(out[l]))))
,0
,255).astype('int16')
# inter-grain scratches
for k in random.sample(range(0,ground.max()), nscratches):
reg = ground==k
if reg.sum() < 500:
continue
c = scipy.ndimage.measurements.center_of_mass(reg)
p = matsci.label.fit_region_z(reg)
w = abs(p[0]-p[2])
h = abs(p[1]-p[3])
largest = int(round(min(w,h)/2))
angle = random.random() * math.pi * 2
val = random.randint(2,48)
for j in range(1,random.randint(1,25)):
l = np.logical_and(
line(
(random.randint(int(c[1]-h/2),
int(c[1]+h/2)),
random.randint(int(c[0]-w/2),
int(c[0]+w/2)))
, angle
, random.randint(2,max(3,largest))
, ground.shape)
, reg)
out[l] += np.clip(
map(int,
map(round,
norm.rvs(
loc=val,
scale=32,
size=len(out[l])))),
0,
255
).astype('int16')
out = np.clip(out,0,255).astype('uint8')
# out = gaussian_filter(out, sigma=1)
# out = median_filter(out, 3)
cv2.imwrite(out_path,out)
return 0
def distort(imgae, config):
""" 向图像中添加噪声
这个函数修改自gqcnn的源程序中，具体原理参考论文
"""
imgae_ = imgae.copy()
# config = self._config
im_height = imgae_.shape[0]
im_width = imgae_.shape[1]
im_center = np.array([float(im_height-1)/2, float(im_width-1)/2])
# denoising and synthetic data generation
if config['multiplicative_denoising']:
gamma_shape = config['gamma_shape']
gamma_scale = 1.0 / gamma_shape
mult_samples = ss.gamma.rvs(gamma_shape, scale=gamma_scale)
imgae_ = imgae_ * mult_samples
# randomly dropout regions of the image for robustness
if config['image_dropout']:
if np.random.rand() < config['image_dropout_rate']:
nonzero_px = np.where(imgae_ > 0)
nonzero_px = np.c_[nonzero_px[0], nonzero_px[1]]
num_nonzero = nonzero_px.shape[0]
num_dropout_regions = ss.poisson.rvs(
config['dropout_poisson_mean'])
# sample ellipses
dropout_centers = np.random.choice(
num_nonzero, size=num_dropout_regions)
x_radii = ss.gamma.rvs(
config['dropout_radius_shape'], scale=config['dropout_radius_scale'], size=num_dropout_regions)
y_radii = ss.gamma.rvs(
config['dropout_radius_shape'], scale=config['dropout_radius_scale'], size=num_dropout_regions)
# set interior pixels to zero
for j in range(num_dropout_regions):
ind = dropout_centers[j]
dropout_center = nonzero_px[ind, :]
x_radius = x_radii[j]
y_radius = y_radii[j]
dropout_px_y, dropout_px_x = sd.ellipse(
dropout_center[0], dropout_center[1], y_radius, x_radius, shape=imgae_.shape)
imgae_[dropout_px_y, dropout_px_x] = 0.0
# dropout a region around the areas of the image with high gradient
if config['gradient_dropout']:
if np.random.rand() < config['gradient_dropout_rate']:
grad_mag = sf.gaussian_gradient_magnitude(
imgae_, sigma=config['gradient_dropout_sigma'])
thresh = ss.gamma.rvs(
config['gradient_dropout_shape'], config['gradient_dropout_scale'], size=1)
high_gradient_px = np.where(grad_mag > thresh)
imgae_[high_gradient_px[0], high_gradient_px[1]] = 0.0
# add correlated Gaussian noise
if config['gaussian_process_denoising']:
gp_rescale_factor = config['gaussian_process_scaling_factor']
gp_sample_height = int(im_height / gp_rescale_factor)
gp_sample_width = int(im_width / gp_rescale_factor)
gp_num_pix = gp_sample_height * gp_sample_width
if np.random.rand() < config['gaussian_process_rate']:
gp_noise = ss.norm.rvs(scale=config['gaussian_process_sigma'], size=gp_num_pix).reshape(
gp_sample_height, gp_sample_width)
# sm.imresize 有警告将被弃用
# gp_noise = sm.imresize(
# gp_noise, gp_rescale_factor, interp='bicubic', mode='F')
# st.resize 用来替用将被弃用的sm.imresize
# gp_noise = st.resize(gp_noise, (im_height, im_width))
gp_noise = cv2.resize(
gp_noise, (im_height, im_width), interpolation=cv2.INTER_CUBIC)
imgae_[imgae_ > 0] += gp_noise[imgae_ > 0]
# run open and close filters to
if config['morphological']:
sample = np.random.rand()
morph_filter_dim = ss.poisson.rvs(
config['morph_poisson_mean'])
if sample < config['morph_open_rate']:
imgae_ = snm.grey_opening(
imgae_, size=morph_filter_dim)
else:
closed_imgae_ = snm.grey_closing(
imgae_, size=morph_filter_dim)
# set new closed pixels to the minimum depth, mimicing the table
new_nonzero_px = np.where(
(imgae_ == 0) & (closed_imgae_ > 0))
closed_imgae_[new_nonzero_px[0], new_nonzero_px[1]] = np.min(
imgae_[imgae_ > 0])
imgae_ = closed_imgae_.copy()
# randomly dropout borders of the image for robustness
if config['border_distortion']:
grad_mag = sf.gaussian_gradient_magnitude(
imgae_, sigma=config['border_grad_sigma'])
high_gradient_px = np.where(
grad_mag > config['border_grad_thresh'])
high_gradient_px = np.c_[
high_gradient_px[0], high_gradient_px[1]]
num_nonzero = high_gradient_px.shape[0]
num_dropout_regions = ss.poisson.rvs(
config['border_poisson_mean'])
# sample ellipses
dropout_centers = np.random.choice(
num_nonzero, size=num_dropout_regions)
x_radii = ss.gamma.rvs(
config['border_radius_shape'], scale=config['border_radius_scale'], size=num_dropout_regions)
y_radii = ss.gamma.rvs(
config['border_radius_shape'], scale=config['border_radius_scale'], size=num_dropout_regions)
# set interior pixels to zero or one
for j in range(num_dropout_regions):
ind = dropout_centers[j]
dropout_center = high_gradient_px[ind, :]
x_radius = x_radii[j]
y_radius = y_radii[j]
dropout_px_y, dropout_px_x = sd.ellipse(
dropout_center[0], dropout_center[1], y_radius, x_radius, shape=imgae_.shape)
if np.random.rand() < 0.5:
imgae_[dropout_px_y, dropout_px_x] = 0.0
else:
imgae_[dropout_px_y, dropout_px_x] = imgae_[
dropout_center[0], dropout_center[1]]
# randomly replace background pixels with constant depth
if config['background_denoising']:
if np.random.rand() < config['background_rate']:
imgae_[imgae_ > 0] = config['background_min_depth'] + (
config['background_max_depth'] - config['background_min_depth']) * np.random.rand()
# symmetrize images
if config['symmetrize']:
# rotate with 50% probability
if np.random.rand() < 0.5:
theta = 180.0
rot_map = cv2.getRotationMatrix2D(
tuple(im_center), theta, 1)
imgae_ = cv2.warpAffine(
imgae_, rot_map, (im_height, im_width), flags=cv2.INTER_NEAREST)
# reflect left right with 50% probability
if np.random.rand() < 0.5:
imgae_ = np.fliplr(imgae_)
# reflect up down with 50% probability
if np.random.rand() < 0.5:
imgae_ = np.flipud(imgae_)
return imgae_
def close_segmentation(segmentation, size, **kwargs):
'''close holes in segmentation maps for training.
'''
return grey_closing(segmentation, size=size, **kwargs)
def main():
# parse cmd arguments
parser = getParser()
parser.parse_args()
args = getArguments(parser)
# prepare logger
logger = Logger.getInstance()
if args.debug: logger.setLevel(logging.DEBUG)
elif args.verbose: logger.setLevel(logging.INFO)
logger.info(
'Executing weighted viscous morphology with {} ({} bins).'.format(
','.join(map(str, args.func)), len(args.func)))
# iterate over input images
for image in args.images:
# build output file name
image_viscous_name = args.folder + '/' + image.split(
'/')[-1][:-4] + '_wviscous_' + '_'.join(map(str, args.func))
image_viscous_name += image.split('/')[-1][-4:]
# check if output file exists
if not args.force:
if os.path.exists(image_viscous_name):
logger.warning(
'The output file {} already exists. Skipping this image.'.
format(image_viscous_name))
continue
# get and prepare image data
logger.info('Loading image {} using NiBabel...'.format(image))
image_gradient = load(image)
# get and prepare image data
image_gradient_data = scipy.squeeze(image_gradient.get_data())
# prepare result image and extract required attributes of input image
if args.debug:
logger.debug(
'Intensity range of gradient image is ({}, {})'.format(
image_gradient_data.min(), image_gradient_data.max()))
# create gradient images flattened histogram
bins = hist_flatened(image_gradient_data, len(args.func))
logger.debug('{} bins created'.format(len(bins) - 1))
# check if the number of bins is consistent
if len(args.func) != len(bins) - 1:
raise Exception(
'Inconsistency between the number of requested and created bins ({} to {})'
.format(args.sections,
len(bins) - 1))
# prepare result file
image_viscous_data = image_gradient_data
# transform the gradient images topography
logger.info(
'Applying the viscous morphological operations on {} sections...'.
format(len(args.func)))
for sl in range(1, len(args.func) + 1):
# create sphere to use in this step
if 0 >= args.func[sl - 1]:
continue # sphere of sizes 0 or below lead to no changes and are not executed
sphere = iterate_structure(generate_binary_structure(3, 1),
args.func[sl - 1]).astype(scipy.int_)
# create masks to extract the affected voxels (i.e. the current slice of the topographic image representation)
mask_greater = (image_gradient_data >= bins[sl]
) # all voxels with are over the current slice
mask_lower = (image_gradient_data < bins[sl - 1]
) # all voxels which are under the current slice
mask_equal = scipy.invert(
mask_greater | mask_lower) # all voxels in the current slice
# extract slice
image_threshold_data = image_gradient_data.copy()
image_threshold_data[
mask_lower] = 0 # set all voxels under the current slice to zero
image_threshold_data[mask_greater] = image_threshold_data[
mask_equal].max(
) # set all voxels over the current slice to the max of all voxels in the current slice
logger.debug('{} of {} voxels belong to this level.'.format(
len(mask_equal.nonzero()[0]),
scipy.prod(image_threshold_data.shape)))
# apply the closing with the appropriate sphere
logger.debug(
'Applying a disk of {} to all values >= {} and < {} (sec {})...'
.format(args.func[sl - 1], bins[sl - 1], bins[sl], sl))
image_closed_data = grey_closing(image_threshold_data,
footprint=sphere)
# add result of this slice to the general results
image_viscous_data = scipy.maximum(image_viscous_data,
image_closed_data)
# save resulting gradient image
logger.info('Saving resulting gradient image as {}...'.format(
image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
logger.info('Successfully terminated.')
m1.saveImage(outputFolder+ timeString() + m1.name + ".png")
x = m.getKmeans(k =k, threshold = threshold)
x['pattern'].saveImage(outputFolder+ getTimeString() + m.name + "threshold%d_clusters%d.png" % (threshold,k))
res[(threshold, k)] = x['pattern']
"""
#########################################
threshold = 40
k = 10
m.load()
m1=m.threshold(0)
m1.show()
m1.matrix = mor.grey_opening(m1.matrix, 5) ####
m1.matrix = mor.grey_closing(m1.matrix, 5) ####
m1.show()
m1.backupMatrix(0)
m1.showWithCoast()
m1.saveImage(outputFolder+m.name+'grey_opening_closing5.png')
m1.restoreMatrix()
threshold=30
x = m1.getKmeans(k =k, threshold = threshold)
x['pattern'].saveImage(outputFolder+ getTimeString() + m.name + "threshold%d_clusters%d.png" % (threshold,k))
threshold=40
m.load()
m.show()
x = m.getKmeans(k =k, threshold = threshold)
def grey_close(im,strel=strel()):
return morphology.grey_closing(im,structure=strel)
frame_disp_count = 0
while motorrunning:
framecount += 1
frame_disp_count += 1
## IMAGE PROCESSING
py_dev.wait_for_frame()
c_im = py_dev.colour
rgb_im = c_im[..., ::-1]
# scaling to map better to color/grayscale
d_im = py_dev.depth * 0.05
# close holes without removing segmentation by doing it before converting to image
d_im_filt = scmorph.grey_closing(d_im, size=(7, 7))
d_im_col = cv2.applyColorMap(d_im.astype(np.uint8),
cv2.COLORMAP_HOT)
# every nth frame, update direction by analysing depth map
# two options: segmentation or gradient. Segmentation is a problem with very noisy images
if framecount > frameint:
yaw_e_prev = yaw_error
# thread.start_new_thread(depthmap_seg_nav, (d_im_col, ))
thread.start_new_thread(depthmap_flow_nav, (d_im_filt, ))
framecount = 1
#if frame_disp_count > frame_disp_int:
# # still super laggy over wifi
# cv2.circle(d_im_col, (cX, cY), 7, (255, 255, 255), -1)
def pre_process_image(img, path=None):
if path: file_name, file_ext = os.path.splitext(os.path.basename(path))
# MORPHOLOGY CLOSING
# http://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.morphology.grey_closing.html#scipy.ndimage.morphology.grey_closing
# http://en.wikipedia.org/wiki/Mathematical_morphology
#
# Cancellare ogni lettera e imperfezione
# Parametri: qui uso il metodo nella sua forma base dove il secondo parametro e' un rettangolo (altezza, larghezza)
# Secondo me teoricamente dovrei matchare la dimesnione media di una parola (una lettera e' troppo piccolo, una riga e' troppo grande)
#
#orig#im = morphology.grey_closing(img, (1, 101))
im = morphology.grey_closing(img, (15, 105)) #odd numbers are better
if path and SAVE_INTERMEDIATE_STEPS: imsave(os.path.join(END_FOLDER, '%s_step1%s' % (file_name, file_ext)), im)
# OTSU THRESHOLDING (statistically optimal)
# http://docs.opencv.org/modules/imgproc/doc/miscellaneous_transformations.html#threshold
# http://en.wikipedia.org/wiki/Otsu%27s_Method
#
# Trasformare l'immagine in due colori: nero per lo sfondo, bianco per il primo piano
# Parametri: 0, 1 (valore per sfondo e primo piano) in produzione
# 0, 255 se volgio vedere l'immagine in bianco e nero per debug
#
#orig#t, im = cv.threshold(im, 0, 1, cv.THRESH_OTSU)
t, im = cv.threshold(im, 0, 255, cv.THRESH_OTSU)
if path and SAVE_INTERMEDIATE_STEPS: imsave(os.path.join(END_FOLDER, '%s_step2%s' % (file_name, file_ext)), im)
# MORPHOLOGY OPENING
# http://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.morphology.grey_opening.html#scipy.ndimage.morphology.grey_opening
# http://en.wikipedia.org/wiki/Mathematical_morphology
#
# Cancellare i sottili bordi bianchi
# Parametri: qui uso il metodo nella sua forma base dove il secondo parametro e' un rettangolo (altezza, larghezza)
# Le dimensioni del rettangolo sono un quadrato di lato = dimesnione minima delpiu grosso extra bordo bianco
#
#origl# im = morphology.grey_opening(im, (51, 51))
im = morphology.grey_opening(im, (51, 51)) #odd numbers are better
if path and SAVE_INTERMEDIATE_STEPS: imsave(os.path.join(END_FOLDER, '%s_step3%s' % (file_name, file_ext)), im)
# CONNECTED-COMPONENT LABELING
# http://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.measurements.label.html#scipy.ndimage.measurements.label
# http://en.wikipedia.org/wiki/Connected-component_labeling
#
# Divido l'immagine (che ora e' teoricamente pulita di tutto il testo) in sotto immagini
# Mantengo solo la sotto immagine piu grossa perche' dovrebbe essere la pagina (il resto lo coloro di nero)
#
# Il risultato puo essere:
# 1 sotto img: caso perfetto, la sotto immagine e' il foglio (il resto e' lo sfondo che non conta, cioe' il bordo nero)
# 2, 3, 4, 5 img: probabilmente c'e' un extra bordo bianco di disturbo su 1,2,3 o 4 lati
# 6+ img: la pagina contiene delle immagini grosso che sono state suddivise
#
# Label divide l'immagine in sottoimmagini e assegna un numero ad ogni pixel (0 e' lo sfondo)
# Tutti i pixel che hanno lo stesso numero fanno parte della stessa sotto immagine
#
# Restituisce:
# una matrice di dimensione uguale all'immagine sorgente dove ogni pixel ha un label
# il numero di sotto immagini identificate
#
lbl, ncc = label(im)
# Identifico la sotto immagine piu grossa
largest = 0, 0
for i in range(1, ncc + 1):
size = len(numpy.where(lbl == i)[0]) #counts how many times the value i is present in the lbl array
if size > largest[1]:
largest = i, size
# Imposto il colore 0 a tutte le sottoimmagini tranne quell apiu grossa
for i in range(1, ncc + 1):
if i == largest[0]:
continue
im[lbl == i] = 0
#Se volessi colorare le sotto immagini in scala di grigi
#import math##
#im[lbl == i] = math.floor(255/ncc-1)*ncc
if path and SAVE_INTERMEDIATE_STEPS: imsave(os.path.join(END_FOLDER, '%s_step4%s' % (file_name, file_ext)), im)###################
return im
def main():
# parse cmd arguments
parser = getParser()
parser.parse_args()
args = getArguments(parser)
# prepare logger
logger = Logger.getInstance()
if args.debug:
logger.setLevel(logging.DEBUG)
elif args.verbose:
logger.setLevel(logging.INFO)
logger.info("Selected viscous type is {}".format(args.type))
# iterate over input images
for image in args.images:
# get and prepare image data
logger.info("Loading image {} using NiBabel...".format(image))
image_gradient = load(image)
# get and prepare image data
image_gradient_data = scipy.squeeze(image_gradient.get_data())
logger.debug(
"Intensity range of gradient image is ({}, {})".format(image_gradient_data.min(), image_gradient_data.max())
)
# build output file name and check for its existence, if not in sections mode
if "sections" != args.type:
# build output file name
image_viscous_name = (
args.folder
+ "/"
+ image.split("/")[-1][:-4]
+ "_viscous_{}_sec_{}_ds_{}".format(args.type, args.sections, args.dsize)
)
image_viscous_name += image.split("/")[-1][-4:]
# check if output file exists
if not args.force:
if os.path.exists(image_viscous_name):
logger.warning("The output file {} already exists. Skipping this image.".format(image_viscous_name))
continue
# execute plain closing i.e. a closing operation over the whole image, if in plain mode
if "plain" == args.type:
# prepare the disc structure (a ball with a diameter of (args.dsize * 2 + 1))
disc = iterate_structure(generate_binary_structure(3, 1), args.dsize).astype(scipy.int_)
# apply closing
logger.info("Applying the morphology over whole image at once...")
image_viscous_data = grey_closing(image_gradient_data, footprint=disc)
# save resulting gradient image
logger.info("Saving resulting gradient image as {}...".format(image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
# skip other morphologies
continue
# create gradient images flattened histogram
bins = hist_flatened(image_gradient_data, args.sections)
logger.debug("{} bins created".format(len(bins) - 1))
# check if the number of bins is consistent
if args.sections != len(bins) - 1:
raise Exception(
"Inconsistency between the number of requested and created bins ({} to {})".format(
args.sections, len(bins) - 1
)
)
# prepare result file
image_viscous_data = image_gradient_data
# transform the gradient images topography (Note: the content of one bin is: bins[slice - 1] <= content < bins[slice]
logger.info("Applying the viscous morphological operations {} times...".format(args.sections))
for slice in range(1, args.sections + 1):
# build output file name and check for its existence, if in sections mode
if "sections" == args.type:
# build output file name
image_viscous_name = (
args.folder
+ "/"
+ image.split("/")[-1][:-4]
+ "_viscous_{}_sec_{}_ds_{}_sl_{}".format(args.type, args.sections, args.dsize, slice)
)
image_viscous_name += image.split("/")[-1][-4:]
# check if output file exists
if not args.force:
if os.path.exists(image_viscous_name):
logger.warning(
"The output file {} already exists. Skipping this slice.".format(image_viscous_name)
)
continue
# prepare result file
image_viscous_data = image_gradient_data
# create masks to extract the affected voxels (i.e. the current slice of the topographic image representation)
mask_greater = image_gradient_data >= bins[slice] # all voxels with are over the current slice
mask_lower = image_gradient_data < bins[slice - 1] # all voxels which are under the current slice
mask_equal = scipy.invert(mask_greater | mask_lower) # all voxels in the current slice
if "mercury" == args.type:
dsize = int((args.dsize / float(args.sections)) * (slice))
disc = iterate_structure(generate_binary_structure(3, 1), dsize).astype(scipy.int_)
mask_equal_or_greater = mask_equal | mask_greater
image_threshold_data = image_gradient_data * mask_equal_or_greater
elif "oil" == args.type:
dsize = int((args.dsize / float(args.sections)) * (args.sections - slice + 1))
disc = iterate_structure(generate_binary_structure(3, 1), dsize).astype(scipy.int_)
image_threshold_data = image_gradient_data.copy()
mask_equal_or_lower = mask_equal | mask_lower
# set all voxels over the current slice to the max of all voxels in the current slice
image_threshold_data[mask_greater] = image_threshold_data[mask_equal_or_lower].max()
elif "sections" == args.type:
dsize = args.dsize
disc = iterate_structure(generate_binary_structure(3, 1), args.dsize).astype(scipy.int_)
image_threshold_data = image_gradient_data.copy()
# set all voxels under the current slice to zero
image_threshold_data[mask_lower] = 0
# set all voxels over the current slice to the max of all voxels in the current slice
image_threshold_data[mask_greater] = image_threshold_data[mask_equal].max()
logger.debug(
"{} of {} voxels belong to this level.".format(
len(mask_equal.nonzero()[0]), scipy.prod(image_threshold_data.shape)
)
)
# apply the closing with the appropriate disc size
logger.debug(
"Applying a disk of {} to all values >= {} and < {}...".format(dsize, bins[slice - 1], bins[slice])
)
image_closed_data = grey_closing(image_threshold_data, footprint=disc)
# add result of this slice to the general results
image_viscous_data = scipy.maximum(image_viscous_data, image_closed_data)
# save created output file, if in sections mode
if "sections" == args.type:
# save resulting gradient image
logger.info("Saving resulting gradient image as {}...".format(image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
# save created output file, if not in sections mode
if "sections" != args.type:
# save resulting gradient image
logger.info("Saving resulting gradient image as {}...".format(image_viscous_name))
image_viscous = image_like(image_viscous_data, image_gradient)
save(image_viscous, image_viscous_name)
logger.info("Successfully terminated.")
pl.save("Stretching", Data)
if(opening):
print "- Morphological Opening..."
Data=grey_opening(Data, structure=Cross)
if(save):
pl.save("Opening", Data)
if(erosion):
print "- Morphological Erosion..."
Data=grey_erosion(Data, structure=Cross)
if(save):
pl.save("Erosion", Data)
if(closing):
print "- Morphological Closing..."
Data=grey_closing(Data, structure=Cross)
if(save):
pl.save("Closing", Data)
# Remark: one could keep on with other transformations, other kernels and so on
# To do so, I would reccomend to use ipython, and eventually load the partial results
FinalStep=Data
if(view):
view_slice(FinalStep,SizeX/2)
pl.show()
print "- Finding maxima..."
local_max=maximum_filter(FinalStep, size=(cutoff/2, cutoff/2, cutoff/2))==FinalStep
Labelled, Num=label(local_max)
print "\n====> Found"+F.RED, Num,F.RESET+"maxima\n"
def orthorectify(args_source_image, args_dsm, args_destination_image,
args_occlusion_thresh=1.0, args_denoise_radius=2,
args_raytheon_rpc=None, args_dtm=None):
"""
Orthorectify an image given the DSM
Args:
source_image: Source image file name
dsm: Digital surface model (DSM) image file name
destination_image: Orthorectified image file name
occlusion-thresh: Threshold on height difference for detecting
and masking occluded regions (in meters)
denoise-radius: Apply morphological operations with this radius
to the DSM reduce speckled noise
raytheon-rpc: Raytheon RPC file name. If not provided
the RPC is read from the source_image
Returns:
COMPLETE_DSM_INTERSECTION = 0
PARTIAL_DSM_INTERSECTION = 1
EMPTY_DSM_INTERSECTION = 2
ERROR = 10
"""
returnValue = COMPLETE_DSM_INTERSECTION
# open the source image
sourceImage = gdal.Open(args_source_image, gdal.GA_ReadOnly)
if not sourceImage:
return ERROR
sourceBand = sourceImage.GetRasterBand(1)
if (args_raytheon_rpc):
# read the RPC from raytheon file
print("Reading RPC from Raytheon file: {}".format(args_raytheon_rpc))
model = raytheon_rpc.read_raytheon_rpc_file(args_raytheon_rpc)
else:
# read the RPC from RPC Metadata in the image file
print("Reading RPC Metadata from {}".format(args_source_image))
rpcMetaData = sourceImage.GetMetadata('RPC')
model = rpc.rpc_from_gdal_dict(rpcMetaData)
if model is None:
print("Error reading the RPC")
return ERROR
# open the DSM
dsm = gdal.Open(args_dsm, gdal.GA_ReadOnly)
if not dsm:
return ERROR
band = dsm.GetRasterBand(1)
dsmRaster = band.ReadAsArray(
xoff=0, yoff=0,
win_xsize=dsm.RasterXSize, win_ysize=dsm.RasterYSize)
dsm_nodata_value = band.GetNoDataValue()
print("DSM raster shape {}".format(dsmRaster.shape))
if args_dtm:
dtm = gdal.Open(args_dtm, gdal.GA_ReadOnly)
if not dtm:
return ERROR
band = dtm.GetRasterBand(1)
dtmRaster = band.ReadAsArray(
xoff=0, yoff=0,
win_xsize=dtm.RasterXSize, win_ysize=dtm.RasterYSize)
newRaster = numpy.where(dsmRaster != dsm_nodata_value, dsmRaster, dtmRaster)
dsmRaster = newRaster
# apply morphology to denoise the DSM
if (args_denoise_radius > 0):
morph_struct = circ_structure(args_denoise_radius)
dsmRaster = morphology.grey_opening(dsmRaster, structure=morph_struct)
dsmRaster = morphology.grey_closing(dsmRaster, structure=morph_struct)
# create the rectified image
driver = dsm.GetDriver()
driverMetadata = driver.GetMetadata()
destImage = None
arrayX = None
arrayY = None
arrayZ = None
if driverMetadata.get(gdal.DCAP_CREATE) == "YES":
print("Create destination image of "
"size:({}, {}) ...".format(dsm.RasterXSize, dsm.RasterYSize))
# georeference information
projection = dsm.GetProjection()
transform = dsm.GetGeoTransform()
gcpProjection = dsm.GetGCPProjection()
gcps = dsm.GetGCPs()
options = ["COMPRESS=DEFLATE"]
# ensure that space will be reserved for geographic corner coordinates
# (in DMS) to be set later
if (driver.ShortName == "NITF" and not projection):
options.append("ICORDS=G")
# If I try to use AddBand with GTiff I get:
# Dataset does not support the AddBand() method.
# So I create all bands using the same type at the begining
destImage = driver.Create(
args_destination_image, xsize=dsm.RasterXSize,
ysize=dsm.RasterYSize,
bands=sourceImage.RasterCount, eType=sourceBand.DataType,
options=options)
if (projection):
# georeference through affine geotransform
destImage.SetProjection(projection)
destImage.SetGeoTransform(transform)
pixels = numpy.arange(0, dsm.RasterXSize)
pixels = numpy.tile(pixels, dsm.RasterYSize)
lines = numpy.arange(0, dsm.RasterYSize)
lines = numpy.repeat(lines, dsm.RasterXSize)
arrayX = transform[0] + pixels * transform[1] + lines * transform[2]
arrayY = transform[3] + pixels * transform[4] + lines * transform[5]
arrayZ = dsmRaster[lines, pixels]
validIdx = arrayZ != dsm_nodata_value
pixels = pixels[validIdx]
lines = lines[validIdx]
arrayX = arrayX[validIdx]
arrayY = arrayY[validIdx]
arrayZ = arrayZ[validIdx]
else:
# georeference through GCPs
destImage.SetGCPs(gcps, gcpProjection)
# not implemented: compute arrayX, arrayY, arrayZ
print("Not implemented yet")
return ERROR
else:
print("Driver {} does not supports Create().".format(driver))
return ERROR
# convert coordinates to Long/Lat
srs = osr.SpatialReference(wkt=projection)
proj_srs = srs.ExportToProj4()
inProj = pyproj.Proj(proj_srs)
outProj = pyproj.Proj('+proj=longlat +datum=WGS84')
arrayX, arrayY = pyproj.transform(inProj, outProj, arrayX, arrayY)
# Sort the points by height so that higher points project last
if (args_occlusion_thresh > 0):
print("Sorting by Height")
heightIdx = numpy.argsort(arrayZ)
arrayX = arrayX[heightIdx]
arrayY = arrayY[heightIdx]
arrayZ = arrayZ[heightIdx]
lines = lines[heightIdx]
pixels = pixels[heightIdx]
# project the points
minZ = numpy.amin(arrayZ)
maxZ = numpy.amax(arrayZ)
# project points to get image indexes and save their height into the image
print("Project {} points to destination image ...".format(len(arrayX)))
print("Points min/max Z: {}/{} ...".format(minZ, maxZ))
print("Projecting Points")
imgPoints = model.project(numpy.array([arrayX, arrayY, arrayZ]).transpose())
intImgPoints = imgPoints.astype(numpy.int).transpose()
# coumpute the bound of the relevant AOI in the source image
print("Source Image size: ", [sourceImage.RasterXSize, sourceImage.RasterYSize])
minPoint = numpy.maximum([0, 0], numpy.min(intImgPoints, 1))
print("AOI min: ", minPoint)
maxPoint = numpy.minimum(numpy.max(intImgPoints, 1),
[sourceImage.RasterXSize,
sourceImage.RasterYSize])
print("AOI max: ", maxPoint)
cropSize = maxPoint - minPoint
if numpy.any(cropSize < 1):
print("DSM does not intersect source image")
returnValue = EMPTY_DSM_INTERSECTION
# shift the projected image point to the cropped AOI space
intImgPoints[0] -= minPoint[0]
intImgPoints[1] -= minPoint[1]
# find indicies of points that fall inside the image bounds
print("Source raster shape {}".format(cropSize))
validIdx = numpy.logical_and.reduce((intImgPoints[1] < cropSize[1],
intImgPoints[1] >= 0,
intImgPoints[0] < cropSize[0],
intImgPoints[0] >= 0))
intImgPoints = intImgPoints[:, validIdx]
# keep only the points that are in the image
numOut = numpy.size(validIdx) - numpy.count_nonzero(validIdx)
if (numOut > 0 and not returnValue == EMPTY_DSM_INTERSECTION):
print("Skipped {} points outside of image".format(numOut))
returnValue = PARTIAL_DSM_INTERSECTION
# use a height map to test for occlusion
if (args_occlusion_thresh > 0):
print("Mapping occluded points")
valid_arrayZ = arrayZ[validIdx]
# render a height map in the source image space
height_map = numpy.full(cropSize[::-1], -numpy.inf, dtype=numpy.float32)
height_map[intImgPoints[1], intImgPoints[0]] = valid_arrayZ
# get a mask of points that locally are (approximately)
# the highest point in the map
is_max_height = height_map[intImgPoints[1], intImgPoints[0]] \
<= valid_arrayZ + args_occlusion_thresh
num_occluded = numpy.size(is_max_height) - numpy.count_nonzero(is_max_height)
print("Skipped {} occluded points".format(num_occluded))
# keep only non-occluded image points
intImgPoints = intImgPoints[:, is_max_height]
# disable occluded points in the valid pixel mask
validIdx[numpy.nonzero(validIdx)[0][numpy.logical_not(is_max_height)]] = False
for bandIndex in range(1, sourceImage.RasterCount + 1):
print("Processing band {} ...".format(bandIndex))
sourceBand = sourceImage.GetRasterBand(bandIndex)
nodata_value = sourceBand.GetNoDataValue()
# for now use zero as a no-data value if one is not specified
# it would probably be better to add a mask (alpha) band instead
if nodata_value is None:
nodata_value = 0
if numpy.any(cropSize < 1):
# read one value for data type
sourceRaster = sourceBand.ReadAsArray(
xoff=0, yoff=0, win_xsize=1, win_ysize=1)
destRaster = numpy.full(
(dsm.RasterYSize, dsm.RasterXSize), nodata_value,
dtype=sourceRaster.dtype)
else:
sourceRaster = sourceBand.ReadAsArray(
xoff=int(minPoint[0]), yoff=int(minPoint[1]),
win_xsize=int(cropSize[0]), win_ysize=int(cropSize[1]))
print("Copying colors ...")
destRaster = numpy.full(
(dsm.RasterYSize, dsm.RasterXSize), nodata_value,
dtype=sourceRaster.dtype)
destRaster[lines[validIdx], pixels[validIdx]] = sourceRaster[
intImgPoints[1], intImgPoints[0]]
print("Write band ...")
destBand = destImage.GetRasterBand(bandIndex)
destBand.SetNoDataValue(nodata_value)
destBand.WriteArray(destRaster)
return returnValue
