def raw2phasecorr(arr_list,clip=0): #cv
import cv2
cx = 0.0
cy = 0.0
stb_arr_list=[]
prev_frame = arr_list[0]
prev_image = np.float32(restoration.denoise_tv_chambolle(prev_frame.astype('uint16'), weight=0.1, multichannel=True)) #ref
for frame in arr_list:
image = np.float32(restoration.denoise_tv_chambolle(frame.astype('uint16'), weight=0.1, multichannel=True))
# TODO: set window around phase correlation
dp = cv2.phaseCorrelate(prev_image, image)
cx = cx - dp[0]
cy = cy - dp[1]
xform = np.float32([[1, 0, cx], [0, 1, cy]])
stable_image = cv2.warpAffine(frame.astype('float32'), xform, dsize=(image.shape[1], image.shape[0]))
prev_image = image
#clip sides
ht,wd=np.shape(stable_image)
# clip=0.125 #0.25
lt=int(wd*clip)
rt=int(wd-wd*clip)
up=int(ht*clip)
dw=int(ht-ht*clip)
stable_image_clipped=stable_image[up:dw,lt:rt]
stb_arr_list.append(stable_image_clipped)
return stb_arr_list
def phasecorr(imlist,imlist2=None,clip=0): #cv [rowini,rowend,colini,colend]
import cv2
cx = 0.0
cy = 0.0
imlist_stb=[]
if imlist2!=None:
imlist2_stb=[]
imi=0
im_prev = imlist[0]
im_denoised_prev = np.float32(restoration.denoise_tv_chambolle(im_prev.astype('uint16'), weight=0.1, multichannel=True)) #ref
for im in imlist:
im_denoised = np.float32(restoration.denoise_tv_chambolle(im.astype('uint16'), weight=0.1, multichannel=True))
# TODO: set window around phase correlation
dp = cv2.phaseCorrelate(im_denoised_prev, im_denoised)
cx = cx - dp[0]
cy = cy - dp[1]
xform = np.float32([[1, 0, cx], [0, 1, cy]])
im_stb = cv2.warpAffine(im.astype('float32'), xform, dsize=(im_denoised.shape[1], im_denoised.shape[0]))
imlist_stb.append(imclipper(im_stb,clip))
if imlist2!=None:
im2=imlist2[imi]
im2_stb=cv2.warpAffine(im2.astype('float32'), xform, dsize=(im_denoised.shape[1], im_denoised.shape[0]))
imlist2_stb.append(imclipper(im2_stb,clip))
im_denoised_prev = im_denoised
imi+=1
if imlist2!=None:
return imlist_stb,imlist2_stb
else:
return imlist_stb
def denoising(astro):
noisy = astro + 0.6 * astro.std() * np.random.random(astro.shape)
noisy = np.clip(noisy, 0, 1)
fig, ax = plt.subplots(nrows=2, ncols=3, figsize=(8, 5), sharex=True,
sharey=True, subplot_kw={'adjustable': 'box-forced'})
plt.gray()
ax[0, 0].imshow(noisy)
ax[0, 0].axis('off')
ax[0, 0].set_title('noisy')
ax[0, 1].imshow(denoise_tv_chambolle(noisy, weight=0.1, multichannel=True))
ax[0, 1].axis('off')
ax[0, 1].set_title('TV')
ax[0, 2].imshow(denoise_bilateral(noisy, sigma_range=0.05, sigma_spatial=15))
ax[0, 2].axis('off')
ax[0, 2].set_title('Bilateral')
ax[1, 0].imshow(denoise_tv_chambolle(noisy, weight=0.2, multichannel=True))
ax[1, 0].axis('off')
ax[1, 0].set_title('(more) TV')
ax[1, 1].imshow(denoise_bilateral(noisy, sigma_range=0.1, sigma_spatial=15))
ax[1, 1].axis('off')
ax[1, 1].set_title('(more) Bilateral')
ax[1, 2].imshow(astro)
ax[1, 2].axis('off')
ax[1, 2].set_title('original')
fig.tight_layout()
plt.show()
def test_denoise_tv_chambolle_multichannel():
denoised0 = restoration.denoise_tv_chambolle(astro[..., 0], weight=0.1)
denoised = restoration.denoise_tv_chambolle(astro, weight=0.1,
multichannel=True)
assert_equal(denoised[..., 0], denoised0)
# tile astronaut subset to generate 3D+channels data
astro3 = np.tile(astro[:64, :64, np.newaxis, :], [1, 1, 2, 1])
# modify along tiled dimension to give non-zero gradient on 3rd axis
astro3[:, :, 0, :] = 2*astro3[:, :, 0, :]
denoised0 = restoration.denoise_tv_chambolle(astro3[..., 0], weight=0.1)
denoised = restoration.denoise_tv_chambolle(astro3, weight=0.1,
multichannel=True)
assert_equal(denoised[..., 0], denoised0)
def plot_preprocessed_image(self):
"""
plots pre-processed image. The plotted image is the same as obtained at the end
of the get_text_candidates method.
"""
image = restoration.denoise_tv_chambolle(self.image, weight=0.1)
thresh = threshold_otsu(image)
bw = closing(image > thresh, square(2))
cleared = bw.copy()
label_image = measure.label(cleared)
borders = np.logical_xor(bw, cleared)
label_image[borders] = -1
image_label_overlay = label2rgb(label_image, image=image)
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(12, 12))
ax.imshow(image_label_overlay)
for region in regionprops(label_image):
if region.area < 10:
continue
minr, minc, maxr, maxc = region.bbox
rect = mpatches.Rectangle((minc, minr), maxc - minc, maxr - minr,
fill=False, edgecolor='red', linewidth=2)
ax.add_patch(rect)
plt.show()
def detect_edges(image_array):
""" Detect edges in a given image
Takes a numpy.array representing an image,
apply filters and edge detection and return a numpy.array
Parameters
----------
image_array : ndarray (2D)
Image data to be processed. Detect edges on this 2D array representing the image
Returns
-------
edges : ndarray (2D)
Edges of an image.
"""
#Transform image into grayscale
img = rgb2gray(image_array)
#Remove some noise from the image
img = denoise_tv_chambolle(img, weight=0.55)
#Apply canny
edges = filter.canny(img, sigma=3.2)
#Clear the borders
clear_border(edges, 15)
#Dilate edges to make them more visible and connected
edges = binary_dilation(edges, selem=diamond(3))
return edges
def compute(self, src):
image = img_as_ubyte(src)
# denoise image
denoised = denoise_tv_chambolle(image, weight=0.05)
denoised_equalize= exposure.equalize_hist(denoised)
# find continuous region (low gradient) --> markers
markers = rank.gradient(denoised_equalize, disk(5)) < 10
markers = ndi.label(markers)[0]
# local gradient
gradient = rank.gradient(denoised, disk(2))
# labels
labels = watershed(gradient, markers)
# display results
fig, axes = plt.subplots(2,3)
axes[0, 0].imshow(image)#, cmap=plt.cm.spectral, interpolation='nearest')
axes[0, 1].imshow(denoised, cmap=plt.cm.spectral, interpolation='nearest')
axes[0, 2].imshow(markers, cmap=plt.cm.spectral, interpolation='nearest')
axes[1, 0].imshow(gradient, cmap=plt.cm.spectral, interpolation='nearest')
axes[1, 1].imshow(labels, cmap=plt.cm.spectral, interpolation='nearest', alpha=.7)
plt.show()
def test_denoise_tv_chambolle_weighting():
# make sure a specified weight gives consistent results regardless of
# the number of input image dimensions
rstate = np.random.RandomState(1234)
img2d = astro_gray.copy()
img2d += 0.15 * rstate.standard_normal(img2d.shape)
img2d = np.clip(img2d, 0, 1)
# generate 4D image by tiling
img4d = np.tile(img2d[..., None, None], (1, 1, 2, 2))
w = 0.2
denoised_2d = restoration.denoise_tv_chambolle(img2d, weight=w)
denoised_4d = restoration.denoise_tv_chambolle(img4d, weight=w)
assert_(measure.compare_ssim(denoised_2d,
denoised_4d[:, :, 0, 0]) > 0.99)
def denoise_image(input, output):
kidney_image = io.imread(input)
# estimate the noise in the image
# do a test denosing using a total variation filter
kidney_image_denoised_tv = restoration.denoise_tv_chambolle(
kidney_image, weight=0.1)
io.imsave(output, kidney_image_denoised_tv)
def blur_predict(model, X, type="median", filter_size=3, sigma=1.0):
if type == "median":
blured_X = np.array(list(map(lambda x: ndimage.median_filter(x, filter_size),
X)))
elif type == "gaussian":
blured_X = np.array(list(map(lambda x: ndimage.gaussian_filter(x, filter_size),
X)))
elif type == "f_gaussian":
blured_X = np.array(list(map(lambda x: filters.gaussian_filter(x.reshape((28, 28)), sigma=sigma).reshape(784),
X)))
elif type == "tv_chambolle":
blured_X = np.array(list(map(lambda x: restoration.denoise_tv_chambolle(x.reshape((28, 28)), weight=0.2).reshape(784),
X)))
elif type == "tv_bregman":
blured_X = np.array(list(map(lambda x: restoration.denoise_tv_bregman(x.reshape((28, 28)), weight=5.0).reshape(784),
X)))
elif type == "bilateral":
blured_X = np.array(list(map(lambda x: restoration.denoise_bilateral(np.abs(x).reshape((28, 28))).reshape(784),
X)))
elif type == "nl_means":
blured_X = np.array(list(map(lambda x: restoration.nl_means_denoising(x.reshape((28, 28))).reshape(784),
X)))
elif type == "none":
blured_X = X
else:
raise ValueError("unsupported filter type", type)
return predict(model, blured_X)
def denoise_image(data, type=None):
from skimage.restoration import denoise_tv_chambolle, denoise_bilateral
if type == "tv":
return denoise_tv_chambolle(data, weight=0.2, multichannel=True)
return denoise_bilateral(data, sigma_range=0.1, sigma_spatial=15)
def preProcessing(imGrayLevel):
#Escopo de algumas Operações básicas utilizadas no pré-processamento:
#..............................................
h = ia.iahistogram(imGrayLevel) #Equalização...
n = imGrayLevel.size
T = 255./n * np.cumsum(h)
T = T.astype(uint8)
#..............................................
T1 = np.arange(256) # função identidade
T2 = ia.ianormalize(np.log(T1+30)) # logaritmica - realce partes escuras
#T5 = ia.ianormalize(T1/50) # reduz o número de níveis de cinza
#..................................................
ax1.imshow(imRGB)
ax1.set_title('rgb')
ax2.imshow(imGrayLevel, vmin=0, vmax=255, cmap=plt.cm.gray)
ax2.set_title('gray level')
imGrayLevel = denoise_tv_chambolle(imGrayLevel, weight=0.1, multichannel=True)
imGrayLevel = img_as_ubyte(imGrayLevel)#Conversão de Float para UINT-8
ax3.imshow(imGrayLevel, vmin=0, vmax=255, cmap=plt.cm.gray) #Filtro de suavização de textura
ax3.set_title('tv signal filter')
realceNucleos = T2[T[imGrayLevel]] #Realce de partes escuras da imagem equalizada
ax4.imshow(realceNucleos, vmin=0, vmax=255, cmap=plt.cm.gray)
ax4.set_title('logaritimica')
return realceNucleos
def test_denoise_tv_chambolle_1d():
"""Apply the TV denoising algorithm on a 1D sinusoid."""
x = 125 + 100 * np.sin(np.linspace(0, 8 * np.pi, 1000))
x += 20 * np.random.rand(x.size)
x = np.clip(x, 0, 255)
res = restoration.denoise_tv_chambolle(x.astype(np.uint8), weight=0.1)
assert_(res.dtype == np.float)
assert_(res.std() * 255 < x.std())
def SaveImage(img, filename):
logging.info('save output to %s', filename)
out = PostprocessImage(img)
if args.remove_noise != 0.0:
out = denoise_tv_chambolle(out,
weight=args.remove_noise,
multichannel=True)
io.imsave(filename, out)
def preprocess(X):
progbar = Progbar(X.shape[0]) # progress bar for pre-processing status tracking
for i in range(X.shape[0]):
for j in range(X.shape[1]):
X[i, j] = denoise_tv_chambolle(X[i, j], weight=0.1, multichannel=False)
progbar.add(1)
return X
def separable_iterated_tv_chambolle(im, sigma_x=1, sigma_y=1, niters=5):
if (sigma_x <= 0) and (sigma_y <= 0):
return im
im_w = np.copy(im)
for i in range(niters):
if sigma_y > 0:
im_w = np.array([
denoise_tv_chambolle(cv,
mad_std(cv) * sigma_y) for cv in im_w.T
]).T # columns
if sigma_x > 0:
im_w = np.array([
denoise_tv_chambolle(rv,
mad_std(rv) * sigma_x) for rv in im_w
]) # rows
return im_w
def test_denoise_tv_chambolle_1d():
"""Apply the TV denoising algorithm on a 1D sinusoid."""
x = 125 + 100*np.sin(np.linspace(0, 8*np.pi, 1000))
x += 20 * np.random.rand(x.size)
x = np.clip(x, 0, 255)
res = restoration.denoise_tv_chambolle(x.astype(np.uint8), weight=0.1)
assert_(res.dtype == np.float)
assert_(res.std() * 255 < x.std())
def denoise_roi(roi, channel=None):
"""Apply further saturation, denoising, unsharp filtering, and erosion
as image preprocessing for blob detection.
Each step can be configured including turned off by
:attr:`config.process_settings`.
Args:
roi: Region of interest as a 3D (z, y, x) array. Note that 4D arrays
with channels are not allowed as the Scikit-Image gaussian filter
only accepts specifically 3 channels, presumably for RGB.
channel (List[int]): Sequence of channel indices in ``roi`` to
saturate. Defaults to None to use all channels.
Returns:
Denoised region of interest.
"""
multichannel, channels = setup_channels(roi, channel, 3)
roi_out = None
for chl in channels:
roi_show = roi[..., chl] if multichannel else roi
settings = config.get_roi_profile(chl)
# find gross density
saturated_mean = np.mean(roi_show)
# further saturation
denoised = np.clip(roi_show, settings["clip_min"],
settings["clip_max"])
tot_var_denoise = settings["tot_var_denoise"]
if tot_var_denoise:
# total variation denoising
denoised = restoration.denoise_tv_chambolle(denoised,
weight=tot_var_denoise)
# sharpening
unsharp_strength = settings["unsharp_strength"]
if unsharp_strength:
blur_size = 8
# turn off multichannel since assume operation on single channel at
# a time and to avoid treating as multichannel if 3D ROI happens to
# have x size of 3
blurred = filters.gaussian(denoised, blur_size, multichannel=False)
high_pass = denoised - unsharp_strength * blurred
denoised = denoised + high_pass
# further erode denser regions to decrease overlap among blobs
thresh_eros = settings["erosion_threshold"]
if thresh_eros and saturated_mean > thresh_eros:
#print("denoising for saturated mean of {}".format(saturated_mean))
denoised = morphology.erosion(denoised, morphology.octahedron(1))
if multichannel:
if roi_out is None:
roi_out = np.zeros(roi.shape, dtype=denoised.dtype)
roi_out[..., chl] = denoised
else:
roi_out = denoised
return roi_out
def TVFilter_I(Im, tv_weight):
""" In-place total variation filter for Image3D's """
Im_np = Im.data.cpu().numpy().squeeze(0)
Im_np = res.denoise_tv_chambolle(Im_np, weight=tv_weight)
Im_data = torch.tensor(Im_np, device=Im.device, dtype=Im.dtype)
Im.data = Im_data.unsqueeze(0).clone()
return Im
def denoise(img_dir, path):
image = io.imread(path)
filename = os.path.splitext(os.path.basename(path))[0]
new_path_bregman = 'denoised_images/orig/bregman/' + img_dir + '/' + filename + '.png'
io.imsave(new_path_bregman, np.clip(denoise_tv_bregman(image, weight=10), -1, 1))
new_path_bregman = 'denoised_images/orig/chambolle/' + img_dir + '/' + filename + '.png'
io.imsave(new_path_bregman, np.clip(denoise_tv_chambolle(image, weight=0.1, multichannel=True), -1, 1))
def denoise(path):
image = io.imread(path)
filename = os.path.splitext(os.path.basename(path))[0]
new_path_bregman = 'denoised_images/bregman/' + filename + '.jpg'
io.imsave(new_path_bregman, denoise_tv_bregman(image, weight=10))
new_path_bregman = 'denoised_images/chambolle/' + filename + '.jpg'
io.imsave(new_path_bregman, denoise_tv_chambolle(image, weight=0.1, multichannel=True))
def preprocess_image(self):
"""
Denoises and increases contrast.
"""
image = restoration.denoise_tv_chambolle(self.image, weight=0.1)
thresh = threshold_otsu(image)
self.bw = closing(image > thresh, square(2))
self.cleared = self.bw.copy()
return self.cleared
def preProcessMLimg(self, image, smallest_size, lowest_region_intensity):
"""
# =======Some preprocessing to get rid of junk parts and make image clearer.=========
#
# --smallest_size/biggest_size: cells size out of this range are ignored.
# --lowest_region_intensity: cells with mean region intensity below this are ignored.
# --cell_region_opening_factor: degree of opening operation on individual cell mask.
# --cell_region_closing_factor: degree of closing operation on individual cell mask.
#====================================================================================
"""
openingfactor=2
closingfactor=3
binary_adaptive_block_size=335
cell_region_opening_factor=1
cell_region_closing_factor=2
image = denoise_tv_chambolle(image, weight=0.01) # Denoise the image.
# -----------------------------------------------Set background to 0-----------------------------------------------
AdaptiveThresholding = threshold_local(image, binary_adaptive_block_size, offset=0)
BinaryMask = image >= AdaptiveThresholding
OpeningBinaryMask = opening(BinaryMask, square(int(openingfactor)))
RegionProposal_Mask = closing(OpeningBinaryMask, square(int(closingfactor)))
clear_border(RegionProposal_Mask)
# label image regions, prepare for regionprops
label_image = label(RegionProposal_Mask)
FinalpreProcessROIMask = np.zeros((image.shape[0], image.shape[1]))
for region in regionprops(label_image,intensity_image = image):
# skip small images
if region.area > smallest_size and region.mean_intensity > lowest_region_intensity:
# draw rectangle around segmented coins
minr, minc, maxr, maxc = region.bbox
bbox_area = (maxr-minr)*(maxc-minc)
# Based on the boundingbox for each cell from first image in the stack, raw image of slightly larger region is extracted from each round.
RawRegionImg = image[minr:maxr, minc:maxc] # Raw region image
#---------Get the cell filled mask-------------
filled_mask_bef, MeanIntensity_Background = imageanalysistoolbox.get_cell_filled_mask(RawRegionImg = RawRegionImg, region_area = bbox_area*0.2,
cell_region_opening_factor = cell_region_opening_factor,
cell_region_closing_factor = cell_region_closing_factor)
filled_mask_convolve2d = imageanalysistoolbox.smoothing_filled_mask(RawRegionImg, filled_mask_bef = filled_mask_bef, region_area = bbox_area*0.2, threshold_factor = 1.1)
#----------Put region maks back to original image.-----------
preProcessROIMask = np.zeros((image.shape[0], image.shape[1]))
preProcessROIMask[minr:maxr, minc:maxc] = filled_mask_convolve2d
FinalpreProcessROIMask += preProcessROIMask
FinalpreProcessROIMask = np.where(FinalpreProcessROIMask > 1, 1, FinalpreProcessROIMask)
ClearedImg = FinalpreProcessROIMask * image
return ClearedImg
def preprocess_image(self):
"""
Denoises and increases contrast.
"""
image = restoration.denoise_tv_chambolle(self.image, weight=0.1)
thresh = threshold_otsu(image)
self.bw = closing(image > thresh, square(2))
self.cleared = self.bw.copy()
return self.cleared
def preprocess_image(self):
# Total-variation denoising
image = restoration.denoise_tv_chambolle(self.image, weight=0.1)
# Return threshold value based on Otsu's method
thresh = threshold_otsu(image)
# Increases contrast
self.bw = closing(image <= thresh, square(1))
self.cleared = self.bw.copy()
return self.cleared
def get_image10():
noisy = denoise_tv_chambolle(io.imread(request.args.get('link')), weight=float(request.args.get('weight')), multichannel=True)
skimage.io.imsave('test_noise.png', noisy)
filename = 'test_noise.png'
return send_file(filename, mimetype='image/gif')
def deoniseTvChambolle():
img = color.rgb2gray(data.astronaut())[:50, :50]
imgO = img.copy()
img += 0.5 * img.std() * np.random.randn(*img.shape)
imgN = img.copy()
denoised_img = denoise_tv_chambolle(img, weight=60)
imgR = denoised_img.copy()
return [imgO, imgN, imgR]
def saveBadScene(scene, weight):
gt, formula = loadScene(scene);
Dd = denoise_tv_chambolle(formula, weight, multichannel=False)
vmin, vmax = np.min(gt), np.max(gt);
plt.figure(figsize=(14,3));
plt.subplot(131); plt.imshow(formula, interpolation='nearest', vmin=vmin, vmax=vmax); plt.title("Baseline"); plt.colorbar(); plt.axis('off');
plt.subplot(132); plt.imshow(Dd, interpolation='nearest', vmin=vmin, vmax=vmax); plt.title("Total Variation"); plt.colorbar(); plt.axis('off');
plt.subplot(133); plt.imshow(gt, interpolation='nearest', vmin=vmin, vmax=vmax); plt.title("Ground Truth %s"%scene); plt.axis('off'); plt.colorbar();
plt.savefig('%s_TV_bad.png'%scene, bbox_inches='tight')
def updateImg(self):
self.Tile = self.Image[self.index[0]:self.index[1],self.index[2]:self.index[3]]
if len(self.Tile.shape)==3:
self.Tile=npy.sum(self.Tile,axis=2)/3
self.Tile = (self.Tile-npy.min(self.Tile))*255.0/(npy.max(self.Tile)-npy.min(self.Tile))
self.Tile = self.Tile.astype(npy.uint8)
self.Tile = npy.transpose(self.Tile,axes=[1,0])
#self.Tile = restoration.denoise_bilateral(self.Tile,win_size=5,sigma_spatial=100,sigma_range=0.5)
self.Tile = restoration.denoise_tv_chambolle(self.Tile, weight=0.1, multichannel=False,n_iter_max=20)
def preprocess(X):
"Pre-process images that are fed to neural network"
progbar = Progbar(X.shape[0]) # progress bar for pre-processing status tracking
for i in range(X.shape[0]):
for j in range(X.shape[1]):
X[i, j] = denoise_tv_chambolle(X[i, j], weight=0.1, multichannel=False)
progbar.add(1)
return X # Denoising weight is the regularization parameter
def _watershed(img, number, region, denoising_weight, sigma, truncate,
min_distance, compactness):
"""
Private function to run the watershed segmentation for a prepared image.
:param img: The image to run the watershed segmentation
:type img: Numpy array
:param number: The unique tile number used for logging
:type number: Integer
:param region: The region for with the watershed segmentation is executed. (Used for logging)
:type region: String
:param denoising_weight: The weight factor for denoising the image before watershed segmentation. See: https://scikit-image.org/docs/dev/api/skimage.restoration.html#skimage.restoration.denoise_tv_chambolle
:type denoising_weight: Float
:param sigma: Sigma value for gaussian blurring the image before watershed segmentation. See: https://scikit-image.org/docs/dev/api/skimage.filters.html#skimage.filters.gaussian
:type sigma: Float
:param truncate: Truncation value for gaussian blurring the image before watershed segmentation. See: https://scikit-image.org/docs/dev/api/skimage.filters.html#skimage.filters.gaussian
:type truncate: Float
:param min_distance: Minimum distance for local maxia representing tree tops.
:type min_distance: Float
:param compactness: Compactness of a watershed basin: See https://scikit-image.org/docs/dev/api/skimage.morphology.html#skimage.morphology.watershed
:type compactness: Float
:return: The image containing the labeled areas for individual trees.
:rtype: Numpy array
"""
if denoising_weight != 0:
denoised = denoise_tv_chambolle(img, weight=denoising_weight)
log.debug("Denoised {} area {}".format(region, number))
else:
denoised = img
# Gauss filter
gauss = gaussian(denoised,
sigma,
mode='constant',
cval=0,
preserve_range=True,
truncate=truncate)
log.debug("Gaussian smoothed {} area {}".format(region, number))
# Create a mask so the segmentation will only occur on the trees
mask = numpy.copy(img)
mask[mask != 0] = 1
# Local maxima
local_max = peak_local_max(gauss,
indices=False,
min_distance=min_distance,
exclude_border=False)
markers = label(local_max)
log.debug("Peaked local maxima in {} area {}".format(region, number))
# watershed
labels = watershed(gauss, markers, mask=mask, compactness=compactness)
log.debug("Applied watershed delineation in {} area {}".format(
region, number))
return labels
def plot_preprocessed_image(self):
'''plots pre-processed image and returns crops of chars.'''
rects = []
image = restoration.denoise_tv_chambolle(self.image, weight=0.2)
thresh = threshold_otsu(image)
bw = closing(image > thresh, square(2))
cleared = bw.copy()
label_image = measure.label(cleared)
#borders = np.logical_xor(bw, cleared)
#label_image[borders] = -1
#image_label_overlay = label2rgb(label_image, image=image)
#fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(12, 12))
#ax.imshow(image_label_overlay)
for region in regionprops(label_image):
if region.area < 10:
continue
minr, minc, maxr, maxc = region.bbox
if minr >= 2:
minr -= 2
if minc >= 2:
minc -= 3
maxr += 1
maxc += 2
rects.append([minc, minr, maxc, maxr]) #EXTREMELY IMPORTANT LINE, [x1, y1, x2, y2]
#rect = mpatches.Rectangle((minc, minr), maxc - minc, maxr - minr,
#fill=False, edgecolor='red', linewidth=2)
#ax.add_patch(rect)
#plt.show()
def is_nested(rects, i, j):
return (rects[j][0] >= rects[i][0] and rects[j][2] <= rects[i][2] and rects[j][1] >= rects[i][1] and rects[j][3] <= rects[i][3]) or (rects[j][0] <= rects[i][0] and rects[j][2] >= rects[i][2] and rects[j][1] <= rects[i][1] and rects[j][3] >= rects[i][3])
# makes list of indices of nested rectangles in rects
deletes = []
for i in range(len(rects)-1):
for j in range(i+1, len(rects)):
if (is_nested(rects, i, j)):
deletes.append(j)
# deletes nested rectangles in rects
num_dels = 0
for d in deletes:
del rects[d-num_dels]
num_dels += 1
print 'successful deletion'
return rects
def reconstruct(img,
drop_rate,
recons,
weight,
drop_rate_post=0,
lab=False,
verbose=False,
input_filepath=''):
assert torch.is_tensor(img)
temp = np.rollaxis(img.numpy(), 0, 3)
w = np.ones_like(temp)
if drop_rate > 0:
# independent channel/pixel salt and pepper
temp2 = random_noise(temp, 's&p', amount=drop_rate, salt_vs_pepper=0)
# per-pixel all channel salt and pepper
r = temp2 - temp
w = (np.absolute(r) < 1e-6).astype('float')
temp = temp + r
if lab:
temp = color.rgb2lab(temp)
if recons == 'none':
temp = temp
elif recons == 'chambolle':
temp = denoise_tv_chambolle(temp, weight=weight, multichannel=True)
elif recons == 'bregman':
if drop_rate == 0:
temp = denoise_tv_bregman(temp, weight=1 / weight, isotropic=True)
else:
temp = minimize_tv_bregman(temp,
w,
weight=1 / weight,
gsiter=10,
eps=0.01,
isotropic=True)
elif recons == 'tvl2':
temp = minimize_tv(temp,
w,
lam=weight,
p=2,
solver='L-BFGS-B',
verbose=verbose)
elif recons == 'tvinf':
temp = minimize_tv_inf(temp,
w,
tau=weight,
p=2,
solver='L-BFGS-B',
verbose=verbose)
else:
print('unsupported reconstruction method ' + recons)
exit()
if lab:
temp = color.lab2rgb(temp)
# temp = random_noise(temp, 's&p', amount=drop_rate_post, salt_vs_pepper=0)
temp = torch.from_numpy(np.rollaxis(temp, 2, 0)).float()
return temp
def test_denoise_tv_chambolle_float_result_range():
# lena image
img = lena_gray
int_lena = np.multiply(img, 255).astype(np.uint8)
assert np.max(int_lena) > 1
denoised_int_lena = restoration.denoise_tv_chambolle(int_lena, weight=60.0)
# test if the value range of output float data is within [0.0:1.0]
assert denoised_int_lena.dtype == np.float
assert np.max(denoised_int_lena) <= 1.0
assert np.min(denoised_int_lena) >= 0.0
def TV_Chambolle(images, factor):
augmented_images = []
for i in range(len(images)):
sigma_est = estimate_sigma(
images[i], multichannel=True, average_sigmas=True) / 100
tv_denoised = denoise_tv_chambolle(images[i], sigma_est * factor)
tv_denoised = (255 * tv_denoised).astype(np.uint8)
augmented_images.append(tv_denoised)
del tv_denoised
return np.asarray(augmented_images)
def denoiseImage(self, sg, thr=1.2):
from skimage.restoration import (denoise_tv_chambolle,
denoise_bilateral, denoise_wavelet,
estimate_sigma)
sigma_est = estimate_sigma(sg, multichannel=False, average_sigmas=True)
sgnew = denoise_tv_chambolle(sg, weight=0.2, multichannel=False)
#sgnew = denoise_bilateral(sg, sigma_color=0.05, sigma_spatial=15, multichannel=False)
#sgnew = denoise_wavelet(sg, multichannel=False)
return sgnew
def tv_chambolle(self, weight=0.1, eps=0.0002, max_iter=200):
denoised = [
R.denoise_tv_chambolle(np.array(item, np.float32),
weight=weight,
eps=eps,
n_iter_max=max_iter,
multichannel=True) for item in self.img
]
return [[cv2.cvtColor(np.array(item, np.uint8), cv2.COLOR_RGB2BGR)]
for item in denoised]
def denoiseTV_Chambolle(imagen,multichannel):
"""
-Tiende a producir imagenes como las de los dibujos animados.
-Reduce al minimo la variacion total de la imagen
"""
noisy = img_as_float(imagen)
denoise = denoise_tv_chambolle(noisy, 7, 9, 0.08,multichannel)
return denoise
def test_denoise_tv_chambolle_float_result_range():
# lena image
img = lena_gray
int_lena = np.multiply(img, 255).astype(np.uint8)
assert np.max(int_lena) > 1
denoised_int_lena = restoration.denoise_tv_chambolle(int_lena, weight=60.0)
# test if the value range of output float data is within [0.0:1.0]
assert denoised_int_lena.dtype == np.float
assert np.max(denoised_int_lena) <= 1.0
assert np.min(denoised_int_lena) >= 0.0
def metric_denoise_tv_chambolle(input, truth):
filtered = denoise_tv_chambolle(input)
filtered = filtered.astype(np.float32)
mse = np.sum((filtered - truth)**2) / input.size
struct_sim = ssim(filtered,
truth,
data_range=filtered.max() - filtered.min())
return mse, struct_sim
def __call__(self, sample):
image, mask = sample
if np.random.rand() > self.prob:
return image, mask
weight = np.random.uniform(low=0, high=self.denoise_weight)
image = denoise_tv_chambolle(image, weight=weight, multichannel=True)
return image, mask
def iterativeBackPropagation(hrImage, lrImages, lrMasks, transforms, H,
itermax, interpOrder):
#Convert LR images to a list of vectors
y = []
for i in range(len(lrImages)):
y.append(
convert_image_to_vector(lrImages[i]) *
convert_image_to_vector(lrMasks[i]))
#Convert HR Image to vector
x = convert_image_to_vector(hrImage)
outputImage = nibabel.Nifti1Image(hrImage.get_data(), hrImage.affine)
#Compute HR mask
hrMaskSum = np.zeros(hrImage.get_data().shape, dtype=np.float32)
for i in range(len(lrImages)):
tmp1 = apply_affine_itk_transform_on_image(input_image=lrMasks[i],
transform=transforms[i][0],
center=transforms[i][1],
reference_image=hrImage,
order=0)
hrMaskSum += tmp1.get_data()
#index = np.nonzero(hrMaskSum)
for j in range(itermax):
error = ibpComputeError(x, H, y,
nibabel.Nifti1Image(hrMaskSum, hrImage.affine),
lrImages, transforms, interpOrder)
# #simulation and error computation
# hrError = np.zeros(hrImage.get_data().shape, dtype=np.float32)
#
# for i in range(len(lrImages)):
# lrError = convert_vector_to_image(H[i].dot(x)-y[i], lrImages[i])
# tmp2 = apply_affine_itk_transform_on_image(input_image = lrError, transform=transforms[i][0], center=transforms[i][1], reference_image=hrImage, order=interpOrder)
# hrError += tmp2.get_data()
#
# hrError2 = np.zeros(hrImage.get_data().shape, dtype=np.float32)
# hrError2[index] = hrError[index] / hrMaskSum[index]
#
#filter error map
from skimage.restoration import denoise_tv_chambolle
hrError2 = denoise_tv_chambolle(error, weight=5)
#update hr image and x
outputImage = nibabel.Nifti1Image(outputImage.get_data() - hrError2,
hrImage.affine)
nibabel.save(nibabel.Nifti1Image(hrError2, hrImage.affine),
'error_iter' + str(j) + '.nii.gz')
nibabel.save(outputImage, 'ibp_iter' + str(j) + '.nii.gz')
x = convert_image_to_vector(outputImage)
return outputImage
def pipeline(image,
shape: tuple = None,
smoothing: float = 0.0,
denoising: float = 0.0,
with_hog_attached: bool = False,
with_dominant_color_attached: bool = False,
pixels_per_cell: tuple = (3, 3),
cells_per_block: tuple = (5, 5),
orientations: int = 9,
converter=None,
debug=False):
if shape and len(shape) > 1:
image = resize(image, shape, anti_aliasing=True, mode='reflect')
original = image
if image.shape[-1] == 4:
image = image[:, :, :3]
if smoothing:
image = gaussian(image, sigma=smoothing, multichannel=False)
if denoising:
image = denoise_tv_chambolle(image,
weight=denoising,
multichannel=True)
converted = None
if callable(converter):
image = converter(image)
converted = image
if with_dominant_color_attached:
multichannel = True if len(
image.shape) == 3 and image.shape[-1] > 1 else False
fd, show = hog(image,
orientations=orientations,
pixels_per_cell=pixels_per_cell,
cells_per_block=cells_per_block,
visualize=True,
multichannel=multichannel,
block_norm="L2-Hys")
dom_color = dominant_color(image, k=3 if multichannel else 2)
if debug:
plot_debug(converted, converter, dom_color, original, shape, show)
image = np.concatenate((dom_color, fd), axis=None)
elif with_hog_attached:
fd, show = hog(image,
orientations=orientations,
pixels_per_cell=pixels_per_cell,
cells_per_block=cells_per_block,
visualize=True,
multichannel=True,
block_norm="L2-Hys")
if debug:
plt.imshow(show)
image = np.concatenate((image.flatten(), fd), axis=None)
image = image.flatten()
return image
def total_variation_filtering(file_path, output_name, weight):
img = img_as_float(io.imread(file_path))
img_denoised = denoise_tv_chambolle(img, weight, multichannel=True)
img_denoised_ubyte = img_as_ubyte(img_denoised)
if len(img_denoised_ubyte.shape) > 2 and img_denoised_ubyte.shape[2] == 4:
img_denoised_ubyte = cv2.cvtColor(img_denoised_ubyte,
cv2.COLOR_BGRA2BGR)
output_folder_path = OUTPUT_FOLDER_NAME + "\\tv_filter\\"
plt.imsave(output_folder_path + output_name, img_denoised_ubyte)
def test_denoise_tv_chambolle_float_result_range():
# astronaut image
img = astro_gray
int_astro = np.multiply(img, 255).astype(np.uint8)
assert_(np.max(int_astro) > 1)
denoised_int_astro = restoration.denoise_tv_chambolle(int_astro,
weight=0.1)
# test if the value range of output float data is within [0.0:1.0]
assert_(denoised_int_astro.dtype == np.float)
assert_(np.max(denoised_int_astro) <= 1.0)
assert_(np.min(denoised_int_astro) >= 0.0)
def test_denoise_tv_chambolle_float_result_range():
# astronaut image
img = astro_gray
int_astro = np.multiply(img, 255).astype(np.uint8)
assert_(np.max(int_astro) > 1)
denoised_int_astro = restoration.denoise_tv_chambolle(int_astro,
weight=0.1)
# test if the value range of output float data is within [0.0:1.0]
assert_(denoised_int_astro.dtype == np.float)
assert_(np.max(denoised_int_astro) <= 1.0)
assert_(np.min(denoised_int_astro) >= 0.0)
def denoise_image(self, img_noise, method='nlm'):
if method == 'nlm':
img_denoised = restoration.denoise_nl_means(
img_noise.reshape(self.data_info['origin_shape'])).reshape(
img_noise.shape)
elif method == 'tv':
img_denoised = restoration.denoise_tv_chambolle(
img_noise.reshape(self.data_info['origin_shape']))
elif method == 'mpo':
img_denoised = self.denoise_image_mpo(img_noise)
return img_denoised
def BF_proc(newimg, backimg, FilterSts, FilterOrder, FilterSelect, back_sbs):
if back_sbs == 2 and FilterSts == 2:
if FilterSelect == 'Median':
ImageFinal = ndimage.median_filter(np.subtract(newimg, backimg),
FilterOrder)
elif FilterSelect == 'Gaussian':
ImageFinal = ndimage.gaussian_filter(np.subtract(newimg, backimg),
FilterOrder)
elif FilterSelect == 'Uniform':
ImageFinal = ndimage.uniform_filter(np.subtract(newimg, backimg),
FilterOrder)
elif FilterSelect == 'Denoise TV':
ImageFinal = denoise_tv_chambolle(np.subtract(newimg, backimg),
weight=FilterOrder,
multichannel=True)
elif FilterSelect == 'Bilateral':
ImageFinal = denoise_bilateral(np.subtract(newimg, backimg),
sigma_range=FilterOrder,
sigma_spatial=15)
elif back_sbs == 2 and FilterSts == 0:
ImageFinal = np.subtract(newimg, backimg)
elif back_sbs == 0 and FilterSts == 2:
if FilterSelect == 'Median':
ImageFinal = ndimage.median_filter(newimg, FilterOrder)
elif FilterSelect == 'Gaussian':
ImageFinal = ndimage.gaussian_filter(newimg, FilterOrder)
elif FilterSelect == 'Uniform':
ImageFinal = ndimage.uniform_filter(newimg, FilterOrder)
elif FilterSelect == 'Denoise TV':
ImageFinal = denoise_tv_chambolle(newimg,
weight=FilterOrder,
multichannel=True)
elif FilterSelect == 'Bilateral':
ImageFinal = denoise_bilateral(newimg,
sigma_range=FilterOrder,
sigma_spatial=15)
elif back_sbs == 0 and FilterSts == 0:
ImageFinal = newimg
return ImageFinal
def test_denoise_tv_chambolle_3d():
"""Apply the TV denoising algorithm on a 3D image representing a sphere."""
x, y, z = np.ogrid[0:40, 0:40, 0:40]
mask = (x - 22)**2 + (y - 20)**2 + (z - 17)**2 < 8**2
mask = 100 * mask.astype(np.float)
mask += 60
mask += 20 * np.random.rand(*mask.shape)
mask[mask < 0] = 0
mask[mask > 255] = 255
res = restoration.denoise_tv_chambolle(mask.astype(np.uint8), weight=0.1)
assert_(res.dtype == np.float)
assert_(res.std() * 255 < mask.std())
def iterativeBackPropagation(hrImage, lrImages, lrMasks, transforms, H, itermax, interpOrder):
# Convert LR images to a list of vectors
y = []
for i in range(len(lrImages)):
y.append(convert_image_to_vector(lrImages[i]) * convert_image_to_vector(lrMasks[i]))
# Convert HR Image to vector
x = convert_image_to_vector(hrImage)
outputImage = nibabel.Nifti1Image(hrImage.get_data(), hrImage.affine)
# Compute HR mask
hrMaskSum = np.zeros(hrImage.get_data().shape, dtype=np.float32)
for i in range(len(lrImages)):
tmp1 = apply_affine_itk_transform_on_image(
input_image=lrMasks[i],
transform=transforms[i][0],
center=transforms[i][1],
reference_image=hrImage,
order=0,
)
hrMaskSum += tmp1.get_data()
# index = np.nonzero(hrMaskSum)
for j in range(itermax):
error = ibpComputeError(
x, H, y, nibabel.Nifti1Image(hrMaskSum, hrImage.affine), lrImages, transforms, interpOrder
)
# #simulation and error computation
# hrError = np.zeros(hrImage.get_data().shape, dtype=np.float32)
#
# for i in range(len(lrImages)):
# lrError = convert_vector_to_image(H[i].dot(x)-y[i], lrImages[i])
# tmp2 = apply_affine_itk_transform_on_image(input_image = lrError, transform=transforms[i][0], center=transforms[i][1], reference_image=hrImage, order=interpOrder)
# hrError += tmp2.get_data()
#
# hrError2 = np.zeros(hrImage.get_data().shape, dtype=np.float32)
# hrError2[index] = hrError[index] / hrMaskSum[index]
#
# filter error map
from skimage.restoration import denoise_tv_chambolle
hrError2 = denoise_tv_chambolle(error, weight=5)
# update hr image and x
outputImage = nibabel.Nifti1Image(outputImage.get_data() - hrError2, hrImage.affine)
nibabel.save(nibabel.Nifti1Image(hrError2, hrImage.affine), "error_iter" + str(j) + ".nii.gz")
nibabel.save(outputImage, "ibp_iter" + str(j) + ".nii.gz")
x = convert_image_to_vector(outputImage)
return outputImage
def test_denoise_tv_chambolle_3d():
"""Apply the TV denoising algorithm on a 3D image representing a sphere."""
x, y, z = np.ogrid[0:40, 0:40, 0:40]
mask = (x - 22)**2 + (y - 20)**2 + (z - 17)**2 < 8**2
mask = 100 * mask.astype(np.float)
mask += 60
mask += 20 * np.random.rand(*mask.shape)
mask[mask < 0] = 0
mask[mask > 255] = 255
res = restoration.denoise_tv_chambolle(mask.astype(np.uint8), weight=0.1)
assert_(res.dtype == np.float)
assert_(res.std() * 255 < mask.std())
def preprocess1(X,weight=0.1):
"""
Pre-process images that are fed to neural network.
:param X: X
"""
progbar = Progbar(X.shape[0]) # progress bar for pre-processing status tracking
for i in range(X.shape[0]):
for j in range(X.shape[1]):
X[i, j] = denoise_tv_chambolle(X[i, j], weight=weight, multichannel=False)
progbar.add(1)
return X
def handle_task(self, gearman_worker, gearman_job):
model_name = (gearman_job.data[:100].split('_')[0]).strip()
data = gearman_job.data[100:]
stream = io.BytesIO(data)
logging.info('model({}) is callbacked'.format(model_name))
start = time.time()
image = self.xp.asarray(Image.open(stream).convert('RGB'),
dtype=self.xp.float32).transpose(2, 0, 1)
image = image.reshape((1,) + image.shape)
x = Variable(image)
if (self.models.has_key(model_name)):
# y = self.model(x)
y = self.models[model_name](x)
else:
return ''
result = cuda.to_cpu(y.data)
result = result.transpose(0, 2, 3, 1)
result = result.reshape((result.shape[1:]))
#result = np.uint8(result)
result = result/255.0
logging.info('{} runtime: {}s'.format(model_name, time.time() - start))
# 增加降噪处理
denoise_rate = self.denoise.get(model_name, self.args.denoise)
logging.info('{} denoise {}'.format(model_name, denoise_rate))
if denoise_rate > 0.001:
denoise_time = time.time()
denoise_rate = 1.0 if denoise_rate>1.0 else denoise_rate
result = denoise_tv_chambolle(result, weight=denoise_rate,
multichannel=True,
n_iter_max = 4)
logging.info('{} denoise runtime {}'.format(model_name, time.time()-denoise_time))
result = np.uint8(result*255)
import ipdb;ipdb.set_trace()
# 读写图片
# file_name = '/tmp/' + str(os.getuid()) + str(int(time.time())) + '.jpg'
file_name = '/tmp/' + str(self.args.gpu) + str(int(time.time() * 10000)) + '.jpg'
Image.fromarray(result).save(file_name)
with open(file_name, 'r') as pf:
content = pf.read()
cmd = 'rm -f {file_name}'.format(file_name=file_name)
commands.getstatusoutput(cmd)
return content
def readChestCTFile(f):
# Read the file into an array
array = np.frombuffer(f.getvalue(), dtype='uint8') # or use uint16?
img = cv2.imdecode(array, cv2.CV_LOAD_IMAGE_GRAYSCALE)
imarray = np.array(img)
imarray = denoise_tv_chambolle(imarray, weight=0.001, multichannel=False)
imarray = (imarray*255).astype('uint8')
slice3D = np.expand_dims(imarray, axis=2)
return slice3D
def findGoldMask(data, options): data = denoise_tv_chambolle(data, weight=0.8, multichannel=False) data = exposure.rescale_intensity(data, out_range=(-1, 1)) thresh = threshold_otsu(data) sigma1 = numpy.std(data[ numpy.where(data<thresh) ]) sigma2 = numpy.std(data[ numpy.where(data>thresh) ]) markers = numpy.zeros(data.shape, dtype=numpy.uint) markers[data < thresh-0.5*sigma1] = 1 markers[data > thresh+0.5*sigma2] = 2 labels = random_walker(data, markers, beta=10, mode='bf') labels[labels != 2]=0 labels[labels == 2]=1 return labels
def test_denoise_tv_chambolle_3d():
"""Apply the TV denoising algorithm on a 3D image representing a sphere."""
x, y, z = np.ogrid[0:40, 0:40, 0:40]
mask = (x - 22) ** 2 + (y - 20) ** 2 + (z - 17) ** 2 < 8 ** 2
mask = 100 * mask.astype(np.float)
mask += 60
mask += 20 * np.random.random(mask.shape)
mask[mask < 0] = 0
mask[mask > 255] = 255
res = restoration.denoise_tv_chambolle(mask.astype(np.uint8), weight=100)
assert res.dtype == np.float
assert res.std() * 255 < mask.std()
# test wrong number of dimensions
assert_raises(ValueError, restoration.denoise_tv_chambolle, np.random.random((8, 8, 8, 8)))
def denoiseMaps(imgs, wt = 0.05, ep = 0.02, cycles = 200):
"""denoiseMaps(imgs, wt = 0.05, ep = 0.02, cycles = 200)
A wrapper for denoise_tv_chambolle
returns
A list of denoised images
"""
from skimage.restoration import denoise_tv_chambolle
denoised = []
for img in imgs:
dn = denoise_tv_chambolle(img, weight=wt, eps=ep, n_iter_max=cycles, multichannel=False)
denoised.append(dn)
return denoised
def run():
print('toto')
import io
import os
import sys
import argparse
from skimage import io
from skimage import filter
from skimage import restoration
from skimage import measure
kidney_image = io.imread('manu.jpg')
# estimate the noise in the image
# do a test denosing using a total variation filter
kidney_image_denoised_tv = restoration.denoise_tv_chambolle( kidney_image, weight=0.1)
io.imsave('denoise_image.jpg', kidney_image_denoised_tv)
def test_denoise_tv_chambolle_2d():
# astronaut image
img = astro_gray.copy()
# add noise to astronaut
img += 0.5 * img.std() * np.random.rand(*img.shape)
# clip noise so that it does not exceed allowed range for float images.
img = np.clip(img, 0, 1)
# denoise
denoised_astro = restoration.denoise_tv_chambolle(img, weight=0.1)
# which dtype?
assert_(denoised_astro.dtype in [np.float, np.float32, np.float64])
from scipy import ndimage as ndi
grad = ndi.morphological_gradient(img, size=((3, 3)))
grad_denoised = ndi.morphological_gradient(denoised_astro, size=((3, 3)))
# test if the total variation has decreased
assert_(grad_denoised.dtype == np.float)
assert_(np.sqrt((grad_denoised**2).sum()) < np.sqrt((grad**2).sum()))
def test_denoise_tv_chambolle_2d():
# lena image
img = lena_gray
# add noise to lena
img += 0.5 * img.std() * np.random.random(img.shape)
# clip noise so that it does not exceed allowed range for float images.
img = np.clip(img, 0, 1)
# denoise
denoised_lena = restoration.denoise_tv_chambolle(img, weight=60.0)
# which dtype?
assert denoised_lena.dtype in [np.float, np.float32, np.float64]
from scipy import ndimage
grad = ndimage.morphological_gradient(img, size=((3, 3)))
grad_denoised = ndimage.morphological_gradient(denoised_lena, size=((3, 3)))
# test if the total variation has decreased
assert grad_denoised.dtype == np.float
assert np.sqrt((grad_denoised ** 2).sum()) < np.sqrt((grad ** 2).sum()) / 2