def updateParametros(val):
global p_segmentos, p_sigma, p_compactness, segments, image, cuda_python
if(val == "Python SLIC"):
cuda_python = 0
elif(val == "CUDA gSLICr"):
cuda_python = 1
p_segmentos = int("%d" % (slider_segmentos.val))
p_sigma = slider_sigma.val
p_compactness = slider_compactness.val
image = c_image.copy()
if(cuda_python == 0):
start_time = time.time()
segments = slic(img_as_float(image), n_segments=p_segmentos, sigma=p_sigma, compactness=p_compactness)
print("--- Tempo Python skikit-image SLIC: %s segundos ---" % (time.time() - start_time))
else:
start_time = time.time()
gSLICrInterface.process( p_segmentos)
print("--- Tempo C++/CUDA gSLICr: %s segundos ---" % (time.time() - start_time))
segments = cuda_seg
obj.set_data(mark_boundaries(img_as_float(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)), segments, outline_color=p_outline))
draw()
def compute_sept_isodata(self, mask, thick, septum_base):
"""Method used to create the cell sept_mask using the threshold_isodata
to separate the cytoplasm from the septum"""
cell_mask = mask
if septum_base:
fluor_box = 1 - self.base_box
else:
fluor_box = self.fluor
perim_mask = self.compute_perim_mask(cell_mask, thick)
inner_mask = cell_mask - perim_mask
inner_fluor = (inner_mask > 0) * fluor_box
threshold = threshold_isodata(inner_fluor[inner_fluor > 0])
interest_matrix = inner_mask * (inner_fluor > threshold)
label_matrix = label(interest_matrix, connectivity=2)
interest_label = 0
interest_label_sum = 0
for l in range(np.max(label_matrix)):
if np.sum(img_as_float(label_matrix == l + 1)) > interest_label_sum:
interest_label = l + 1
interest_label_sum = np.sum(img_as_float(label_matrix == l + 1))
return img_as_float(label_matrix == interest_label)
def createFigure4(list_file, list_param, corpus_meta, prob_topic_doc, segment_dir, n_topic, output_dir):
for file_item in list_file:
print file_item
for topic in range(n_topic):
print topic
fig = plt.figure()
img = img_as_float(io.imread(img_dir+file_item+'.ppm'))
ax1 = fig.add_subplot(4,4, 1, axisbg='grey')
ax1.set_xticks(()), ax1.set_yticks(())
ax1.imshow(img)
index=2
for param in list_param:
if 'slic' in param:
# print 'test', index
# print corpus_meta[0][1].split('-')[0]
segment_in_file = [item_corpus for item_corpus in corpus_meta if file_item == item_corpus[1].split('-')[0]]
print len(segment_in_file)
segments_res = csv2Array(segment_dir+'/'+file_item+'/'+file_item+'-'+param+'.sup')
img = img_as_float(io.imread(img_dir+file_item+'.ppm'))
output = np.zeros( (len(img), len(img[0])) )
for segment in segment_in_file:
# print prob_topic_doc[int(segment[0])][topic]
output[segments_res == int(segment[2])] = prob_topic_doc[int(segment[0])][topic]
output = mark_boundaries(output, segments_res)
ax1 = fig.add_subplot(4,4, index, axisbg='grey')
# ax1 = fig.add_subplot(5,10, index, axisbg='grey')
ax1.set_xticks(()), ax1.set_yticks(())
ax1.imshow(output)
index += 1
ensure_path(output_dir+'/'+file_item+'/')
plt.savefig(output_dir+'/'+file_item+'/topic-'+str(topic)+'-'+file_item+'.pdf')
plt.clf()
plt.close()
def get(self, uri):
i = imread(uri)
if len(i.shape) == 2:
i = gray2rgb(i)
else:
i = i[:, :, :3]
c = self._image_to_color.get(i)
dbg = self._settings['debug']
if dbg is None:
return c
c, imgs = c
b = splitext(basename(uri))[0]
imsave(join(dbg, b + '-resized.jpg'), imgs['resized'])
imsave(join(dbg, b + '-back.jpg'), img_as_float(imgs['back']))
imsave(join(dbg, b + '-skin.jpg'), img_as_float(imgs['skin']))
imsave(join(dbg, b + '-clusters.jpg'), imgs['clusters'])
return c, {
'resized': join(dbg, b + '-resized.jpg'),
'back': join(dbg, b + '-back.jpg'),
'skin': join(dbg, b + '-skin.jpg'),
'clusters': join(dbg, b + '-clusters.jpg'),
}
def slic_data():
for i in range(uu_num_train+uu_num_test):
print "data %d" %(i+1)
img_name = ''
if i < 10:
img_name = '0' + str(i)
else:
img_name = str(i)
#Read first 70 images as floats
img = img_as_float(io.imread('..\data\\training\image_2\uu_0000' + img_name + '.png'))
img_hsv = color.rgb2hsv(img)
gt_img = img_as_float(io.imread('..\data\\training\gt_image_2\uu_road_0000' + img_name + '.png'))
#Create superpixels for training images
image_segment = slic(img, n_segments = numSegments, sigma = 5)
t, train_indices = np.unique(image_segment, return_index=True)
images_train_indices.append(train_indices)
image = np.reshape(img,(1,(img.shape[0]*img.shape[1]),3))
image_hsv = np.reshape(img_hsv,(1,(img_hsv.shape[0]*img_hsv.shape[1]),3))
#images_train.append([image[0][i] for i in train_indices])
images_train_hsv.append([image_hsv[0][i] for i in train_indices])
#Create gt training image values index at train_indices and converted to 1 or 0
gt_image = np.reshape(gt_img, (1,(gt_img.shape[0]*gt_img.shape[1]),3))
gt_image = [1 if gt_image[0][i][2] > 0 else 0 for i in train_indices]
gt_images_train.append(gt_image)
def compute_mask(self, params):
"""Creates the mask for the base image.
Needs the base image, an instance of imageloaderparams
and the clip area, which should be already defined
by the load_base_image method.
Creates the mask by improving the base mask created by the
compute_base_mask method. Applies the mask closing, dilation and
fill holes parameters.
"""
self.compute_base_mask(params)
mask = np.copy(self.base_mask)
closing_matrix = np.ones((params.mask_closing, params.mask_closing))
if params.mask_closing > 0:
# removes small dark spots and then small white spots
mask = img_as_float(morphology.closing(
mask, closing_matrix))
mask = 1 - \
img_as_float(morphology.closing(
1 - mask, closing_matrix))
for f in range(params.mask_dilation):
mask = morphology.erosion(mask, np.ones((3, 3)))
if params.mask_fill_holes:
# mask is inverted
mask = 1 - img_as_float(ndimage.binary_fill_holes(1.0 - mask))
self.mask = mask
self.overlay_mask_base_image()
def svm_predict(case,svm_classifier):
print "svm predict"
A = []
b = []
if case == 1:
for i in range(uu_num_test):
img_name = i + uu_num_train
if img_name < 10:
img_name = '0' + str(img_name)
else:
img_name = str(img_name)
print "data %d " %(i+uu_num_train)
#Read test images as floats
img = img_as_float(io.imread('..\data\\training\image_2\uu_0000' + img_name + '.png'))
gt_img = img_as_float(io.imread('..\data\\training\gt_image_2\uu_road_0000' + img_name + '.png'))
#Create superpixels for test images
image_segment = slic(img, n_segments = numSegments, sigma = 5)
t, train_indices = np.unique(image_segment, return_index=True)
images_train_indices.append(train_indices)
image = np.reshape(img,(1,(img.shape[0]*img.shape[1]),3))
images_train.append([image[0][i] for i in train_indices])
#Create gt test image values index at train_indices and converted to 1 or 0
gt_image = np.reshape(gt_img, (1,(gt_img.shape[0]*gt_img.shape[1]),3))
gt_image = [1 if gt_image[0][i][2] > 0 else 0 for i in train_indices]
print "len of gt_image: %d with ones: %d" %(len(gt_image),gt_image.count(1))
gt_images_train.append(gt_image)
for i in range(uu_num_test):
for j in range(len(images_train[i])):
A.append(images_train[i][j])
b.append(gt_images_train[i][j])
else:
for i in range(uu_num_train,uu_num_train+uu_num_test):
for j in range(len(images_train_hsv[i])):
#val = np.zeros(6)
#val[0:3] = images_train[i][j]
#val[3:6] = images_train_hsv[i][j]
#A.append(val)
#A.append(images_train[i][j])
A.append(images_train_hsv[i][j])
b.append(gt_images_train[i][j])
A = np.asarray(A)
b = np.asarray(b)
print "A.shape = %s, b.shape = %s" %(A.shape,b.shape)
predicted = svm_classifier.predict(A)
print svm_classifier.score(A,b)
#for i in range(len(gt_images_train)):
# for j in range(len(gt_images_train[i])):
# print "%d, %d" %(gt_images_train[i][j], predicted[j])
print("Classification report for classifier %s:\n%s\n" %(svm_classifier,metrics.classification_report(b,predicted)))
def run_metrics(result_dir, metrics_initial_path, noisy_image_dir):
# Add initial metric values
csv_file = open(metrics_initial_path, 'r')
csv_reader = csv.DictReader(csv_file)
metrics_initial = {}
for row in csv_reader:
metrics_initial[row['image']] = {
'q': float(row['q']),
'ocr': float(row['ocr']),
'mse': float(row['mse'])
}
csv_file.close()
results = []
DEFAULT_TEXT = 'Lorem ipsum\ndolor sit amet,\nconsectetur\n\nadipiscing elit.\n\nDonec vel\naliquet velit,\nid congue\nposuere.'
runs = list(xrange(1, 10 + 1, 1))
for run in runs:
# print "--- RUN " + run
run_dir = os.path.join(result_dir, str(run))
for filename in os.listdir(run_dir):
if not filename.endswith('.png'):
continue
image_name = os.path.splitext(filename)[0]
blur, noise, contrast = _parse_image_name(image_name)
image_path = os.path.join(run_dir, filename)
image = util.img_as_float(io.imread(image_path))
ocr = ocr_accuracy(image, DEFAULT_TEXT)
try:
result = next(
r for r in results
if r['image']['name'] == image_name)
except StopIteration:
result = {
'image': {
'name': image_name,
'blur': blur,
'noise': noise,
'contrast': contrast
},
'metrics_initial': metrics_initial[image_name],
'ocr': [],
'q': [],
'mse': [],
}
results.append(result)
result['ocr'].append(ocr)
result['q'].append(q_py(image))
# MSE
ideal_image_name = 'noisy-00-00-' + str(contrast).replace('.', '') + '.png'
ideal_image_path = os.path.join(noisy_image_dir, ideal_image_name)
ideal_image = util.img_as_float(io.imread(ideal_image_path))
mse_val = mse(ideal_image, image)
result['mse'].append(mse_val)
return results
def get_images(noisy_image_dir, clear_image_dir):
images = []
for noisy_image_file in sorted(os.listdir(noisy_image_dir)):
name, ext = os.path.splitext(noisy_image_file)
contrast = name.split('-')[::-1][0]
clear_image_name = 'clear-00-00-' + contrast + ext
noisy_image_path = os.path.join(noisy_image_dir, noisy_image_file)
clear_image_path = os.path.join(clear_image_dir, clear_image_name)
# Read images
noisy_image = util.img_as_float(io.imread(noisy_image_path))
clear_image = util.img_as_float(io.imread(clear_image_path))
images.append((noisy_image, clear_image, name))
return images
def create_bin(img, otsu_method=True):
# Binary image created from Threshold, then labelling is done on this image
if otsu_method:
int_img = rescale(img)
t_otsu = threshold_otsu(int_img)
bin_img = (int_img >= t_otsu)
float_img = img_as_float(bin_img)
return float_img
else:
thresh = 400
int_img = rescale(img)
bin_img = (int_img >= thresh)
float_img = img_as_float(bin_img)
return float_img
def showPredictionOutput(self):
image_withText = self.image.copy()
# show the output of the prediction with text
for (i, segVal) in enumerate(np.unique(self.segments)):
CORD = self.centerList[i]
if self.predictionList[i] == "other":
colorFont = (255, 0, 0) # "Blue color for other"
else:
colorFont = (0, 0, 255) # "Red color for ocean"
#textOrg = CORD
#textOrg = tuple(numpy.subtract((10, 10), (4, 4)))
testOrg = (40,40) # need this for the if statment bellow
# for some yet unknown reason CORD does sometime contain somthing like this [[[210 209]] [[205 213]] ...]
# the following if statemnet is to not get a error becouse of this
if len(CORD) == len(testOrg):
#textOrg = tuple(np.subtract(CORD, (12, 0)))
textOrg = CORD
cv2.putText(self.image, self.predictionList[i], textOrg, cv2.FONT_HERSHEY_SIMPLEX, 0.1, colorFont, 3)
markedImage = mark_boundaries(img_as_float(cv2.cvtColor(self.image, cv2.COLOR_BGR2RGB)), self.segments)
else:
pass
cv2.imshow("segmented image", markedImage)
cv2.waitKey(0)
def harris_ones(img, window_size, k=0.05):
"""Calculate the harris score based on a window function of diagonal ones.
Args:
img The image to use for corner detection.
window_size Size of the window (NxN).
k Weighting parameter during the final scoring (det vs. trace).
Returns:
Corner score image
"""
# Gradients
img = skiutil.img_as_float(img)
imgy, imgx = np.gradient(img)
imgxy = imgx * imgy
imgxx = imgx ** 2
imgyy = imgy ** 2
# window function (matrix of diagonal ones)
window = np.ones((window_size, window_size))
# compute parts of harris matrix
a11 = signal.correlate(imgxx, window, mode="same") / window_size
a12 = signal.correlate(imgxy, window, mode="same") / window_size
a21 = a12
a22 = signal.correlate(imgyy, window, mode="same") / window_size
# compute score per pixel
det_a = a11 * a22 - a12 * a21
trace_a = a11 + a22
return det_a - k * trace_a ** 2
def get_image(fname):
arr = io.imread(fname)
if arr.ndim == 2:
arr = color.gray2rgb(arr)
arr = util.img_as_float(arr)
assert arr.ndim == 3
return arr
def segment_selection_using_GA(name_file, param):
global map_segment
map_segment = generate_map_segment(name_file, param)
global img
img = img_as_float(io.imread(param['img_dir']+name_file+'.ppm'))
ind_segment = do_selection(name_file)
create_segmentfile_from_binary_string(name_file, param['segment_dir'], param['corpus_meta'], param['output_dir'], param['img_dir'], ind_segment)
def plot_top_N_segment_in_topic(list_file, segment_dir, corpus_meta, prob_topic_doc, n_topic, N, output_dir):
index = 0
for name_file in list_file:
index += 1
print name_file, index
for topic in range(n_topic):
fig = plt.figure()
index_plt = 1
ind_segment = [int(item_corpus[0]) for item_corpus in corpus_meta if name_file == item_corpus[1].split('-')[0]]
val_segment = [prob_topic_doc[ind][topic] for ind in ind_segment]
ind_of_ind_segment = get_index_bestdoc(val_segment, 12)
for index in ind_of_ind_segment:
segments_res = csv2Array(segment_dir+'/'+name_file+'/'+ corpus_meta[ind_segment[index]][1])
no_segment = int(corpus_meta[ind_segment[index]][2])
output = img_as_float(io.imread(img_dir+name_file+'.ppm'))
output[segments_res != no_segment] = (0.0,0.0,0.9)
ax1 = fig.add_subplot(4,3, index_plt, axisbg='grey')
ax1.set_xticks(()), ax1.set_yticks(())
ax1.set_title(prob_topic_doc[ind_segment[index]][topic])
ax1.imshow(output)
index_plt += 1
ensure_path(output_dir+'/'+name_file+'/')
plt.savefig(output_dir+'/'+name_file+'/'+str(topic)+'.pdf')
plt.clf()
plt.close()
def main():
"""Load template image (needle) and the large example image (haystack),
generate matching score per pixel, plot the results."""
img_haystack = skiutil.img_as_float(data.camera()) # the image in which to search
img_needle = img_haystack[140:190, 220:270] # the template to search for
img_sad = np.zeros(img_haystack.shape) # score image
height_h, width_h = img_haystack.shape
height_n, width_n = img_needle.shape
# calculate score for each pixel
# stop iterating over pixels when the whole template cannot any more (i.e. stop
# at bottom and right border)
for y in range(height_h - height_n):
for x in range(width_h - width_n):
patch = img_haystack[y:y+height_n, x:x+width_n]
img_sad[y, x] = sad(img_needle, patch)
img_sad = img_sad / np.max(img_sad)
# add highest score to bottom and right borders
img_sad[height_h-height_n:, :] = np.max(img_sad[0:height_h, 0:width_h])
img_sad[:, width_h-width_n:] = np.max(img_sad[0:height_h, 0:width_h])
# plot results
util.plot_images_grayscale(
[img_haystack, img_needle, img_sad],
["Image", "Image (Search Template)", "Matching (darkest = best match)"]
)
def get_saliency_ft(img_path): # Saliency map calculation based on: img = skimage.io.imread(img_path) img_rgb = img_as_float(img) img_lab = skimage.color.rgb2lab(img_rgb) mean_val = np.mean(img_rgb,axis=(0,1)) kernel_h = (1.0/16.0) * np.array([[1,4,6,4,1]]) kernel_w = kernel_h.transpose() blurred_l = scipy.signal.convolve2d(img_lab[:,:,0],kernel_h,mode='same') blurred_a = scipy.signal.convolve2d(img_lab[:,:,1],kernel_h,mode='same') blurred_b = scipy.signal.convolve2d(img_lab[:,:,2],kernel_h,mode='same') blurred_l = scipy.signal.convolve2d(blurred_l,kernel_w,mode='same') blurred_a = scipy.signal.convolve2d(blurred_a,kernel_w,mode='same') blurred_b = scipy.signal.convolve2d(blurred_b,kernel_w,mode='same') im_blurred = np.dstack([blurred_l,blurred_a,blurred_b]) sal = np.linalg.norm(mean_val - im_blurred,axis = 2) sal_max = np.max(sal) sal_min = np.min(sal) sal = 255 * ((sal - sal_min) / (sal_max - sal_min)) return sal
def create_best_segment_image(list_file, segment_dir, corpus_meta, prob_topic_doc, n_topic, n_segment, output_dir):
for name_file in list_file:
print name_file
buff_segment = []
ind_segment = [int(item_corpus[0]) for item_corpus in corpus_meta if name_file == item_corpus[1].split('-')[0]]
val_segment = [(ind, prob_topic_doc[ind][topic]) for ind in ind_segment for topic in range(n_topic)]
val_segment = sorted(val_segment,key=itemgetter(1), reverse=True)
bag_of_best_segment = val_segment[0:n_segment]
fig = plt.figure()
index_plt = 1
for index, prob in bag_of_best_segment:
print 'jogie'
img = img_as_float(io.imread(img_dir+name_file+'.ppm'))
n_col = len(img[0])
list_of_segment = csvToListOfSegment(segment_dir+'/'+name_file+'/'+ corpus_meta[index][1])
no_segment = int(corpus_meta[index][2])
buff_segment.append([row*n_col+col for row, col in list_of_segment[no_segment]])
segments_res = csv2Array(segment_dir+'/'+name_file+'/'+ corpus_meta[index][1])
img[segments_res != no_segment] = (0.0,0.0,0.9)
ax1 = fig.add_subplot(6,5, index_plt, axisbg='grey')
ax1.set_xticks(()), ax1.set_yticks(())
ax1.imshow(img)
index_plt += 1
plt.savefig('coba/'+name_file)
plt.clf()
plt.close()
def enumerate_images(image_dir, format='png'):
for filename in os.listdir(image_dir):
if os.path.splitext(filename)[1].endswith(format):
image_path = os.path.join(image_dir, filename)
image = util.img_as_float(io.imread(image_path))
image_name = os.path.splitext(filename)[0]
yield image_name, image
def run_initial(noisy_image_dir, output_file):
DEFAULT_TEXT = 'Lorem ipsum\ndolor sit amet,\nconsectetur\n\nadipiscing elit.\n\nDonec vel\naliquet velit,\nid congue\nposuere.'
fieldnames = ['image', 'q', 'ocr', 'mse']
csv_file = open(output_file, 'w')
writer = csv.DictWriter(csv_file, fieldnames=fieldnames)
writer.writeheader()
clear_images = _read_clear_images()
for noisy_image_file in sorted(os.listdir(noisy_image_dir)):
if noisy_image_file.endswith('.png'):
image = util.img_as_float(io.imread(
os.path.join(noisy_image_dir, noisy_image_file)))
image_name = os.path.splitext(noisy_image_file)[0]
# All three metrics
q = q_py(image)
ocr = ocr_accuracy(image, DEFAULT_TEXT)
_, __, contrast = _parse_image_name(image_name)
contrast_str = str(contrast).replace('.', '')
clear_image_name = 'clear-00-00-' + contrast_str
mse_val = mse(clear_images[clear_image_name], image)
result_row = {
'image': image_name,
'q': q,
'ocr': ocr,
'mse': mse_val
}
writer.writerow(result_row)
csv_file.close()
return None
def _compute_auto_correlation(image, sigma):
"""Compute auto-correlation matrix using sum of squared differences.
Parameters
----------
image : ndarray
Input image.
sigma : float
Standard deviation used for the Gaussian kernel, which is used as
weighting function for the auto-correlation matrix.
Returns
-------
Axx : ndarray
Element of the auto-correlation matrix for each pixel in input image.
Axy : ndarray
Element of the auto-correlation matrix for each pixel in input image.
Ayy : ndarray
Element of the auto-correlation matrix for each pixel in input image.
"""
if image.ndim == 3:
image = img_as_float(rgb2grey(image))
imx, imy = _compute_derivatives(image)
# structure tensore
Axx = ndimage.gaussian_filter(imx * imx, sigma, mode='constant', cval=0)
Axy = ndimage.gaussian_filter(imx * imy, sigma, mode='constant', cval=0)
Ayy = ndimage.gaussian_filter(imy * imy, sigma, mode='constant', cval=0)
return Axx, Axy, Ayy
def run_q(result_dir, output_file):
csv_file = open(output_file, 'w')
# fieldnames = ['image', 'param_set', 'final_error', 'training_time']
fieldnames = ['image', 'param_set', 'q']
writer = csv.DictWriter(csv_file, fieldnames=fieldnames)
writer.writeheader()
# Enumerate images
for image_name in sorted(os.listdir(result_dir)):
image_dir = os.path.join(result_dir, image_name)
if os.path.isdir(image_dir):
print image_name
# Enumerate parameter sets
for result_file in sorted(os.listdir(image_dir)):
if result_file.endswith('.png'):
image_path = os.path.join(image_dir, result_file)
image = util.img_as_float(io.imread(image_path))
q = q_py(image)
# result_json_path = os.path.join(image_dir, result_json)
result_ps_name = os.path.splitext(result_file)[0]
# result_data = read_results(result_json_path)
# # last_epoch_error = float(parse_epochs(result_data)[-1]['error'])
# last_epoch_error = float(parse_epochs(result_data)[-1])
# # Write into csv file
result_row = {
'image': image_name,
'param_set': result_ps_name,
'q': q,
}
writer.writerow(result_row)
csv_file.close()
return None
def run_ocr(result_dir, output_file):
DEFAULT_TEXT = 'Lorem ipsum\ndolor sit amet,\nconsectetur\n\nadipiscing elit.\n\nDonec vel\naliquet velit,\nid congue\nposuere.'
csv_file = open(output_file, 'w')
fieldnames = ['image', 'param_set', 'ocr']
writer = csv.DictWriter(csv_file, fieldnames=fieldnames)
writer.writeheader()
# Enumerate images
for image_name in sorted(os.listdir(result_dir)):
image_dir = os.path.join(result_dir, image_name)
if os.path.isdir(image_dir):
print image_name
# Enumerate parameter sets
for result_file in sorted(os.listdir(image_dir)):
if result_file.endswith('.png'):
image_path = os.path.join(image_dir, result_file)
image = util.img_as_float(io.imread(image_path))
ocr = ocr_accuracy(image, DEFAULT_TEXT)
result_ps_name = os.path.splitext(result_file)[0]
# # Write into csv file
result_row = {
'image': image_name,
'param_set': result_ps_name,
'ocr': ocr,
}
writer.writerow(result_row)
csv_file.close()
return None
def SegmentationFelz_run_2d(rod):
img = img_as_float(
RodriguesToUnambiguousColor(rod["x"], rod["y"], rod["z"], maxRange=None, centerR=None).astype("uint8")
)
segments_slic = felzenszwalb(img, scale=100, sigma=0.0, min_size=10)
print("Slic number of segments: %d" % len(np.unique(segments_slic)))
return segments_slic
def to_norm(arr):
"""
Helper function to normalise/scale an array. This is needed for example
for scikit-image which uses floats between 0 and 1.
Parameters
----------
arr : `~numpy.ndarray`
Array to normalise.
Returns
-------
arr : `~numpy.ndarray`
Array with values between 0 (min) and 1 (max)
Examples
--------
>>> import numpy as np
>>> from sunpy.image.util import to_norm
>>> out = to_norm(np.array([-1, 0, 1]))
>>> out
array([ 0. , 0.5, 1. ])
"""
arr = np.array(arr, dtype='double')
arr = img_as_float(arr, force_copy=True)
if arr.min() < 0:
arr += np.abs(arr.min())
arr /= arr.max()
return arr
def build_target(args): target = img_as_float(io.imread(args['target'])) target = color.rgb2gray(target) ratio = float(target.shape[0]) / float(target.shape[1]) target = transform.resize(target, (sz, int(sz/ratio))) return target
def segment(self):
self.segments = slic(img_as_float(self.img), enforce_connectivity=True)
self.mask = np.zeros(self.img.shape[:2],dtype='int' )
self.mask = self.mask - 1
for (i, segVal) in enumerate(np.unique(self.segments)):
self.mask[segVal == self.segments] = i
self.pixel_list.append(self.Pixel(i))
def recreate_images(result_dir, noisy_image_dir):
# Read noisy images first
test_images = {}
for image_name in os.listdir(noisy_image_dir):
if image_name.endswith('.png'):
image_path = os.path.join(noisy_image_dir, image_name)
image = util.img_as_float(io.imread(image_path))
image_name_noext = os.path.splitext(image_name)[0]
test_images[image_name_noext] = image
# Enumerate results - image directories
for image_name in sorted(os.listdir(result_dir)):
image_dir = os.path.join(result_dir, image_name)
if os.path.isdir(image_dir):
print image_name
for result_file in sorted(os.listdir(image_dir)):
if result_file.endswith('.net'):
# Instantiate trained ANN from .net file
net_path = os.path.join(image_dir, result_file)
ann = libfann.neural_net()
ann.create_from_file(net_path)
# Filter the same image which it was trained with
filtered_image = filter_fann(
test_images[image_name], ann)
param_set_name = os.path.splitext(result_file)[0]
io.imsave(
os.path.join(image_dir, param_set_name + '.png'),
filtered_image)
def harris_gauss(img, sigma=1, k=0.05):
"""Calculate the harris score based on a gauss window function.
Args:
img The image to use for corner detection.
sigma The sigma value for the gauss functions.
k Weighting parameter during the final scoring (det vs. trace).
Returns:
Corner score image"""
# Gradients
img = skiutil.img_as_float(img)
imgy, imgx = np.gradient(img)
imgxy = imgx * imgy
imgxx = imgx ** 2
imgyy = imgy ** 2
# compute parts of harris matrix
a11 = ndimage.gaussian_filter(imgxx, sigma=sigma, mode="constant")
a12 = ndimage.gaussian_filter(imgxy, sigma=sigma, mode="constant")
a21 = a12
a22 = ndimage.gaussian_filter(imgyy, sigma=sigma, mode="constant")
# compute score per pixel
det_a = a11 * a22 - a12 * a21
trace_a = a11 + a22
score = det_a - k * trace_a ** 2
return score
def superpixels(image):
""" given an input image, create super pixels on it
"""
# we could try to limit the problem of holes in boundary by first over segmenting the image
import matplotlib.pyplot as plt
from skimage.segmentation import felzenszwalb, slic, quickshift
from skimage.segmentation import mark_boundaries
from skimage.util import img_as_float
jac_float = img_as_float(image)
plt.imshow(jac_float)
#segments_fz = felzenszwalb(jac_float, scale=100, sigma=0.5, min_size=50)
segments_slic = slic(jac_float, n_segments=600, compactness=0.01, sigma=0.001
#, multichannel = False
, max_iter=50)
fig, ax = plt.subplots(1, 1, sharex=True, sharey=True, subplot_kw={'adjustable':'box-forced'})
fig.set_size_inches(8, 3, forward=True)
fig.subplots_adjust(0.05, 0.05, 0.95, 0.95, 0.05, 0.05)
#ax[0].imshow(mark_boundaries(jac, segments_fz))
#ax[0].set_title("Felzenszwalbs's method")
ax.imshow(mark_boundaries(jac_float, segments_slic))
ax.set_title("SLIC")
return segments_slic
def adjust_contast_brightness(image: np.ndarray,
alpha=1.0,
betta=0.0) -> np.ndarray:
image = img_as_float(image)
return np.clip(alpha * image + betta, 0, 1)
def process_wrapper(self, **kwargs):
"""
From scikit-image: Quick shift segments image using quickshift clustering in Color-(x,y) space.
Real time : No
Keyword Arguments (in parentheses, argument name):
* Color space (color_space): Selected color space, default: RGB
* Width of Gaussian kernel (kernel_size): Width of Gaussian kernel used in smoothing the sample density. Higher means fewer clusters.
* Max distance (max_dist): Cut-off point for data distances. Higher means fewer clusters
* Ratio (ratio): Balances color-space proximity and image-space proximity. Higher values give more weight to color-space.
"""
wrapper = self.init_wrapper(**kwargs)
if wrapper is None:
return False
color_space = self.get_value_of("color_space")
kernel_size = self.get_value_of("kernel_size")
max_dist = self.get_value_of("max_dist")
ratio = self.get_value_of("ratio") / 100
res = False
try:
img = self.wrapper.current_image
if color_space.upper() == "HSV":
img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
elif color_space.upper() == "LAB":
img = cv2.cvtColor(img, cv2.COLOR_BGR2LAB)
if kernel_size == 1:
kernel_size = 0
elif kernel_size % 2 == 0:
kernel_size += 1
img = img_as_float(img)
labels = quickshift(img,
kernel_size=kernel_size,
max_dist=max_dist,
ratio=ratio)
self.result = labels.copy()
labels[labels == -1] = 0
labels = ((labels - labels.min()) / (labels.max() - labels.min()) *
255).astype(np.uint8)
water_img = cv2.applyColorMap(255 - labels, DEFAULT_COLOR_MAP)
wrapper.store_image(water_img, "quick_shift_vis")
self.demo_image = self.print_segmentation_labels(
watershed_image=water_img.copy(),
labels=labels,
dbg_suffix="quick_shift",
)
except Exception as e:
res = False
wrapper.error_holder.add_error(
new_error_text=f'Failed to process {self. name}: "{repr(e)}"',
new_error_level=35,
target_logger=logger,
)
else:
res = True
finally:
return res
def detector(frame,
average,
roi,
output,
cluster_pars=["dbscan", 10, 10],
threshold_pars=["otsu"],
total_frames=False,
fit_kind="circle",
block_shape=(256, 256),
nthreads=2,
regex="[0-9]{6,}",
debug=False):
"""
Detect whitecapping using a local thresholding approach.
There is no output, this function will write to file.
Parameters:
----------
frame : str
Full frame path
averages : str
Full average path.
Use compute_average_image.py to obtain the averages.
roi : list or pd.DataFrame
Region of interest for processing.
Use minmun_bounding_geometry.py to obtain a valid file.
output : str
Output path for the processed data.
cluster_pars : list
List of parameters for clustering.
threshold_pars : list
List of parameters for thresholding.
total_frames: int
Total number of frames. for plotting only
fit_kind : str
What geometry to fit to a cluster. Can be either circle or ellipse.
block_shape : tupple
Block size for view_as_blocks. must be a power of 2
regex : string
Regex to get the sequential image number "[0-9]{6,}".
debug : bool
Run in debug mode. will run in serial and plot outputs.
Returns:
-------
Nothing. Will write to file instead.
"""
# try to figure out frame number and process ID
PID = os.getpid()
frmid = int(re.search(regex, os.path.basename(frame)).group())
print(" -- started processing frame", frmid, "of", total_frames,
"with PID", PID)
# ---- try the detection pipeline ----
try:
# read
img = img_as_float(rgb2gray(imread(frame)))
# ---- deal with roi ----
try:
roi_coords, roi_rect, mask = compute_roi(roi, frame, regex=regex)
except Exception:
print(" -- died because of Region of Interest processing frame",
frmid)
return 0
# try to read average image
try:
avg = img_as_float(rgb2gray(imread(average)))
except Exception:
avg = np.zeros(img.shape)
# remove average and mask where the intensity decreases
dif = img - avg
dif[dif < 0] = 0
mask = rgb2gray(mask)
# TODO: Verify effect of scalling instead of subtraction
dif = img_as_ubyte(dif)
img = img_as_ubyte(img)
avg = img_as_ubyte(avg)
# threshold
if threshold_pars[0] == "otsu":
trx = otsu_threshold(dif)
bin_img = apply_threshold(dif * mask, trx)
bin_img = np.invert(bin_img)
elif threshold_pars[0] == "entropy":
kptrx = kapur_threshold(dif)
bin_img = apply_threshold(dif * mask, kptrx)
bin_img = np.invert(bin_img)
elif threshold_pars[0] == "adaptative":
local_thresh = threshold_local(dif * mask,
threshold_pars[1],
offset=threshold_pars[2])
bin_img = dif > local_thresh
elif threshold_pars[0] == "constant":
bin_img = apply_threshold(dif * mask, int(threshold_pars[1]))
bin_img = np.invert(bin_img)
elif threshold_pars[0] == "file":
trxdf = pd.read_csv(threshold_pars[1])
nearest = trxdf["frame"].values[np.argmin(
np.abs(frmid - trxdf["frame"].values))]
bin_img = apply_threshold(dif * mask,
trxdf.iloc[nearest]["threshold"])
bin_img = np.invert(bin_img)
else:
raise ValueError("Fatal: could not deal with thresholding method.")
# ensure the shape is right for processing as blocks
bin_img, block_shape = ensure_shape(bin_img, block_shape)
view = view_as_blocks(bin_img, tuple(block_shape.tolist()))
# outputs
dfs = [] # store dbscan results
# loop over image blocks
for i in range(view.shape[0]):
for j in range(view.shape[1]):
# target block
blk = view[i, j, :, :]
# update indexes
i1 = block_shape[0] * i
i2 = i1 + block_shape[0]
j1 = block_shape[1] * j
j2 = j1 + block_shape[1]
# try to group bright pixels
try:
df = cluster(np.invert(blk),
cluster_pars[1],
cluster_pars[2],
backend=cluster_pars[0],
nthreads=nthreads,
fit_kind=fit_kind)
if not df.empty:
# fix offsets
# i, j need to swaped here, not sure why
df["ic"] = df["ic"] + j1
df["jc"] = df["jc"] + i1
# add info about the processing blocks
df["block_i"] = j
df["block_j"] = i
df["block_i_left"] = j1
df["block_i_right"] = j2
df["block_j_top"] = i1
df["block_j_bottom"] = i2
# append
dfs.append(df)
except Exception:
raise
pass # do nothing here
# it means that a block search failled
# concatenate
if dfs:
df_dbscan = pd.concat(dfs)
# add some extra information
# df_dbscan["step"] = int(os.path.basename(frame).split(".")[0])
df_dbscan["frame"] = frmid
# write to file
fname = str(frmid).zfill(8) + ".csv"
df_dbscan.to_csv(os.path.join(output, fname), index=False)
else:
print(" -- no clusters were found processing frame", frmid)
df_dbscan = pd.DataFrame()
return 0
# debug plot
if debug:
fig, ax = plot(frmid, img, bin_img, block_shape, roi_rect,
df_dbscan, total_frames, fit_kind)
# save to file
fname = str(frmid).zfill(8) + ".png"
plt.savefig(os.path.join(output, fname),
dpi=150,
bbox_inches="tight",
pad_inches=0.1)
# plt.show()
plt.close()
except Exception:
raise
print(" -- died for some unknown reason processing frame", frmid)
print(" -- finished processing frame", frmid)
return 1
def vein_gabor_enhancement(img, scale):
I = img_as_float(img)
No_of_Orientation = 16 # in wrist paper 16 oreientations
No_of_Scale = len(scale)
a = scale
Gr, _ = Gabor(0, 1.5, 1, 200, a[No_of_Scale-1]) # 1.5
p = Energy_check(Gr)
Gabor_Matrix_Gr = np.zeros(
(2*p + 1, 2*p + 1, No_of_Orientation, No_of_Scale))
ER = [0] * No_of_Scale
for s in range(No_of_Scale):
ang_count = 0
for ang in range(0, 179, int(np.ceil(180 / No_of_Orientation))):
Gr, _ = Gabor(ang, 1.5, 1, p, a[s]) # 1.5
Gabor_Matrix_Gr[:, :, ang_count, s] = Gr
ang_count = ang_count + 1
if ang_count == No_of_Orientation:
break
R = Energy_check(np.squeeze(Gabor_Matrix_Gr[:, :, 1, s]))
ER[s] = R
Energy_Map = np.zeros(I.shape)
Scale_Map = np.ones(I.shape)
Orient_Map = np.ones(I.shape)
E = np.zeros((I.shape[0], I.shape[1], No_of_Orientation, No_of_Scale))
Gabor_fft_Matrix_Gr = np.zeros(
(I.shape[0], I.shape[1], No_of_Orientation, No_of_Scale))
I_fft = np.fft.fft2(I)
for s in range(No_of_Scale):
for ang in range(No_of_Orientation):
pad1 = math.floor((I.shape[0] - (2*p+1)) / 2)
pad2 = math.floor((I.shape[1] - (2*p+1)) / 2)
pad = [(pad1, pad2), (pad1, pad2)]
Gabor_fft_Matrix_Gr[:, :, ang, s] = np.fft.fft2(
np.pad(Gabor_Matrix_Gr[:, :, ang, s], pad, mode='constant'), I.shape)
I2 = np.square(I)
Image_Power = np.zeros((I.shape[0], I.shape[1], No_of_Scale))
for s in range(No_of_Scale):
Image_Power[:, :, s] = np.sqrt(
signal.convolve2d(I2, np.ones((ER[s], ER[s])), 'same')) + 0.1
for ang in range(No_of_Orientation):
E[:, :, ang, s] = np.fft.fftshift(
np.fft.ifft2(-Gabor_fft_Matrix_Gr[:, :, ang, s] * I_fft)) / Image_Power[:, :, s]
EngAng = np.zeros((I.shape[0], I.shape[1], No_of_Scale))
AngIdx = np.zeros((I.shape[0], I.shape[1], No_of_Scale))
for s in range(No_of_Scale):
EngAng[:, :, s] = np.amax(E[:, :, :, s], axis=2)
AngIdx[:, :, s] = np.argmax(E[:, :, :, s], axis=2)
Energy_Map = np.amax(EngAng, axis=2)
Scale_Map = np.argmax(EngAng, axis=2)
for x in range(Energy_Map.shape[0]):
for y in range(Energy_Map.shape[1]):
Orient_Map[x, y] = AngIdx[x, y, Scale_Map[x, y]]
return Orient_Map
def test_li_coins_image_as_float():
coins = util.img_as_float(data.coins())
assert 94 / 255 < threshold_li(coins) < 95 / 255
def test_equalize_float():
img = util.img_as_float(test_img)
img_eq = exposure.equalize_hist(img)
cdf, bin_edges = exposure.cumulative_distribution(img_eq)
check_cdf_slope(cdf)
def paint_segment_skimage(self,
image,
color,
px=0,
py=0,
idx_segment=[],
border=True,
clear=False):
"""Paint a list of segments using a index or position in image.
Parameters
----------
image : opencv 3-channel color image.
Segmented image.
color : string
X11Color name.
px : integer, optional, default = 0
Segment point inside the image in x-axis.
py : integer, optional, default = 0
Segment point inside the image in y-axis.
idx_segment : list, optional, default = []
List of segments.
border : boolean, optional, default = True
If true paint the border of segments with default color.
clear : boolean, optional, default = False
If true clear previous painting in the image.
Returns
-------
new_image : opencv 3-channel color image.
Painted image.
run_time : integer
Running time spent in milliseconds.
"""
# Check if segmentation was already performed
if self._segments is None:
return image, 0
start_time = TimeUtils.get_time()
# If idx_segment not passed, get the segment index using the points (px, py)
if len(idx_segment) == 0:
idx_segment = [self._segments[py, px]]
height, width, channels = self._original_image.shape
# Create a mask, painting black all pixels outside of segments and white the pixels inside
mask_segment = np.zeros(self._original_image.shape[:2], dtype="uint8")
for idx in idx_segment:
mask_segment[self._segments == idx] = 255
mask_inv = cv2.bitwise_not(mask_segment)
# Paint all pixels in original image with choosed color
class_color = np.zeros((height, width, 3), np.uint8)
class_color[:, :] = X11Colors.get_color(color)
if SkimageSegmenter.FORCE_OPT == False:
colored_image = cv2.addWeighted(self._original_image, 0.7,
class_color, 0.3, 0)
else:
colored_image = cv2.addWeighted(image, 0.7, class_color, 0.3, 0)
colored_image = cv2.bitwise_and(colored_image,
colored_image,
mask=mask_segment)
# Create a new image keeping the painting only in pixels inside of segments
new_image = image if clear == False else self._original_image
new_image = cv2.bitwise_and(new_image, new_image, mask=mask_inv)
mask_segment[:] = 255
new_image = cv2.bitwise_or(new_image, colored_image, mask=mask_segment)
# If border is true, paint the segments borders
if border == True and SkimageSegmenter.FORCE_OPT == False:
color = X11Colors.get_color_zero_one(
self.border_color.get_cast_val())
outline_color = color if self.border_outline.value == 'Yes' else None
new_image = img_as_ubyte(
mark_boundaries(img_as_float(new_image),
self._segments.astype(np.int8),
color=color,
outline_color=outline_color))
end_time = TimeUtils.get_time()
# Return painted image
return new_image, (end_time - start_time)
Created on Tue Jun 20 09:23:45 2017
@author: viper
"""
matplotlib.use('pgf') # Force Matplotlib back-end
# Modules....
import matplotlib
import numpy as np
import matplotlib.pyplot as plt
from skimage import io
from scipy import ndimage
from skimage import segmentation
from skimage.util import img_as_float
from skimage import color
plt.close('all') # Close all remaining figures
filename = 'DoP.tiff'
im = io.imread(filename) # Open the image
im = img_as_float(im)
im = color.grey2rgb(im)
plt.figure(1)
plt.imshow(im, cmap='gray')
labels = segmentation.felzenszwalb(im, scale=800, sigma=0.5, min_size=250)
plt.imshow(segmentation.mark_boundaries(color.label2rgb(labels, im), labels))
plt.savefig('Processed/Felzenszwalb/' + filename)
of Quickshift, while ``n_segments`` chooses the number of centers for kmeans.
.. [3] Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi,
Pascal Fua, and Sabine Suesstrunk, SLIC Superpixels Compared to
State-of-the-art Superpixel Methods, TPAMI, May 2012.
"""
import matplotlib.pyplot as plt
import numpy as np
from skimage.data import lena
from skimage.segmentation import felzenszwalb, slic, quickshift
from skimage.segmentation import mark_boundaries
from skimage.util import img_as_float
img = img_as_float(lena()[::2, ::2])
segments_fz = felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
segments_slic = slic(img, ratio=10, n_segments=250, sigma=1)
segments_quick = quickshift(img, kernel_size=3, max_dist=6, ratio=0.5)
print("Felzenszwalb's number of segments: %d" % len(np.unique(segments_fz)))
print("Slic number of segments: %d" % len(np.unique(segments_slic)))
print("Quickshift number of segments: %d" % len(np.unique(segments_quick)))
fig, ax = plt.subplots(1, 3)
fig.set_size_inches(8, 3, forward=True)
plt.subplots_adjust(0.05, 0.05, 0.95, 0.95, 0.05, 0.05)
ax[0].imshow(mark_boundaries(img, segments_fz))
ax[0].set_title("Felzenszwalbs's method")
ax[1].imshow(mark_boundaries(img, segments_slic))
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image", required=True, help="Path to the image")
args = vars(ap.parse_args())
# load the image and apply SLIC and extract (approximately)
# the supplied number of segments
image = cv2.imread(args["image"])
image = cv2.resize(image, (800, 600))
segments = slic(image, n_segments=40)
# show the output of SLIC
fig = plt.figure("Superpixels")
ax = fig.add_subplot(1, 1, 1)
ax.imshow(
mark_boundaries(img_as_float(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)),
segments))
plt.axis("off")
plt.show()
# loop over the unique segment values
for (i, segVal) in enumerate(np.unique(segments)):
# construct a mask for the segment
print "[x] inspecting segment %d" % (i)
mask = np.zeros(image.shape[:2], dtype="uint8")
mask[segments == segVal] = 255
# show the masked region
cv2.imshow("Applied", cv2.bitwise_and(image, image, mask=mask))
cv2.waitKey(0)
def adaptiveEqualization(img: np.ndarray,
clip_limit: float = 0.03) -> np.ndarray:
img_f = img_as_float(img)
logging.info(img_f[0, :4])
return img_as_ubyte(exposure.equalize_adapthist(img_f, clip_limit))
# In[ ]:
# In[50]:
# 3 segmentation algorithms: felzenszwalb, slic, quickshift
from __future__ import print_function
import matplotlib.pyplot as plt
import numpy as np
from skimage.segmentation import felzenszwalb, slic, quickshift
from skimage.segmentation import mark_boundaries
from skimage.util import img_as_float
img = img_as_float(resize(image, (100, 100)))
# img = img_as_float(resize(image, (50, 50)))
segments_fz = felzenszwalb(img, scale=100, sigma=0.5, min_size=50)
segments_slic = slic(img, n_segments=250, compactness=10, sigma=1)
segments_quick = quickshift(img, kernel_size=3, max_dist=6, ratio=0.5)
print("Felzenszwalb's number of segments: %d" % len(np.unique(segments_fz)))
print("Slic number of segments: %d" % len(np.unique(segments_slic)))
print("Quickshift number of segments: %d" % len(np.unique(segments_quick)))
fig, ax = plt.subplots(1, 3)
fig.set_size_inches(8, 3, forward=True)
fig.subplots_adjust(0.05, 0.05, 0.95, 0.95, 0.05, 0.05)
ax[0].imshow(mark_boundaries(img, segments_fz))
return min_i
# prepare filter bank kernels
kernels = []
for theta in range(4):
theta = theta / 4. * np.pi
for sigma in (1, 3):
for frequency in (0.05, 0.25):
kernel = np.real(gabor_kernel(frequency, theta=theta,
sigma_x=sigma, sigma_y=sigma))
kernels.append(kernel)
shrink = (slice(0, None, 3), slice(0, None, 3))
brick = img_as_float(data.load('brick.png'))[shrink]
grass = img_as_float(data.load('grass.png'))[shrink]
wall = img_as_float(data.load('rough-wall.png'))[shrink]
image_names = ('brick', 'grass', 'wall')
images = (brick, grass, wall)
# prepare reference features
ref_feats = np.zeros((3, len(kernels), 2), dtype=np.double)
ref_feats[0, :, :] = compute_feats(brick, kernels)
ref_feats[1, :, :] = compute_feats(grass, kernels)
ref_feats[2, :, :] = compute_feats(wall, kernels)
print('Rotated images matched against references using Gabor filter banks:')
print('original: brick, rotated: 30deg, match result: ', end='')
feats = compute_feats(ndi.rotate(brick, angle=190, reshape=False), kernels)
def carregaImg(arg):
return data.imread("../imagens/" + arg)
#carrega imagem
img_1 = carregaImg(sys.argv[1])
nome_img_saida = sys.argv[2]
mask_size = int(sys.argv[3])
# Se necessário, converte para imagem em níveis de cinza.
if img_1.ndim > 2:
img_1 = color.rgb2gray(img_1)
# Converte imagem para float.
img_1 = util.img_as_float(img_1)
#processa a imagem
masc_random = np.ones([mask_size, mask_size], dtype=float)
masc_random = masc_random / float(mask_size ^ 2)
img = img_1
img_1 = filters.convolve(img_1, masc_random, mode='constant', cval=0)
# Operadores de Sobel
sob_h = np.array([[-1., -2., -1.], [0., 0., 0.], [1., 2., 1.]], dtype=float)
sob_v = np.array([[-1., 0., 1.], [-2., 0., 2.], [-1., 0., 1.]], dtype=float)
# Gradiente de Sobel.
im_sob_h = filters.convolve(img_1, sob_h)
im_sob_v = filters.convolve(img_1, sob_v)
def random_walker(data,
labels,
mode='bf',
tol=1.e-3,
copy=True,
return_full_prob=False,
spacing=None,
alpha=0.3,
beta=0.3,
gamma=0.4,
a=130.0,
b=10.0,
c=800.0):
"""Random walker algorithm for segmentation from markers.
Random walker algorithm is implemented for gray-level or multichannel
images.
Parameters
----------
data : array_like
Image to be segmented in phases. Gray-level `data` can be two- or
three-dimensional; multichannel data can be three- or four-
dimensional (multichannel=True) with the highest dimension denoting
channels. Data spacing is assumed isotropic unless the `spacing`
keyword argument is used.
labels : array of ints, of same shape as `data` without channels dimension
Array of seed markers labeled with different positive integers
for different phases. Zero-labeled pixels are unlabeled pixels.
Negative labels correspond to inactive pixels that are not taken
into account (they are removed from the graph). If labels are not
consecutive integers, the labels array will be transformed so that
labels are consecutive. In the multichannel case, `labels` should have
the same shape as a single channel of `data`, i.e. without the final
dimension denoting channels.
beta : float
Penalization coefficient for the random walker motion
(the greater `beta`, the more difficult the diffusion).
mode : string, available options {'cg_mg', 'cg', 'bf'}
Mode for solving the linear system in the random walker algorithm.
If no preference given, automatically attempt to use the fastest
option available ('cg_mg' from pyamg >> 'cg' with UMFPACK > 'bf').
- 'bf' (brute force): an LU factorization of the Laplacian is
computed. This is fast for small images (<1024x1024), but very slow
and memory-intensive for large images (e.g., 3-D volumes).
- 'cg' (conjugate gradient): the linear system is solved iteratively
using the Conjugate Gradient method from scipy.sparse.linalg. This is
less memory-consuming than the brute force method for large images,
but it is quite slow.
- 'cg_mg' (conjugate gradient with multigrid preconditioner): a
preconditioner is computed using a multigrid solver, then the
solution is computed with the Conjugate Gradient method. This mode
requires that the pyamg module (http://pyamg.org/) is
installed. For images of size > 512x512, this is the recommended
(fastest) mode.
tol : float
tolerance to achieve when solving the linear system, in
cg' and 'cg_mg' modes.
copy : bool
If copy is False, the `labels` array will be overwritten with
the result of the segmentation. Use copy=False if you want to
save on memory.
multichannel : bool, default False
If True, input data is parsed as multichannel data (see 'data' above
for proper input format in this case)
return_full_prob : bool, default False
If True, the probability that a pixel belongs to each of the labels
will be returned, instead of only the most likely label.
spacing : iterable of floats
Spacing between voxels in each spatial dimension. If `None`, then
the spacing between pixels/voxels in each dimension is assumed 1.
Returns
-------
output : ndarray
* If `return_full_prob` is False, array of ints of same shape as
`data`, in which each pixel has been labeled according to the marker
that reached the pixel first by anisotropic diffusion.
* If `return_full_prob` is True, array of floats of shape
`(nlabels, data.shape)`. `output[label_nb, i, j]` is the probability
that label `label_nb` reaches the pixel `(i, j)` first.
See also
--------
skimage.morphology.watershed: watershed segmentation
A segmentation algorithm based on mathematical morphology
and "flooding" of regions from markers.
Notes
-----
Multichannel inputs are scaled with all channel data combined. Ensure all
channels are separately normalized prior to running this algorithm.
The `spacing` argument is specifically for anisotropic datasets, where
data points are spaced differently in one or more spatial dimensions.
Anisotropic data is commonly encountered in medical imaging.
The algorithm was first proposed in *Random walks for image
segmentation*, Leo Grady, IEEE Trans Pattern Anal Mach Intell.
2006 Nov;28(11):1768-83.
The algorithm solves the diffusion equation at infinite times for
sources placed on markers of each phase in turn. A pixel is labeled with
the phase that has the greatest probability to diffuse first to the pixel.
The diffusion equation is solved by minimizing x.T L x for each phase,
where L is the Laplacian of the weighted graph of the image, and x is
the probability that a marker of the given phase arrives first at a pixel
by diffusion (x=1 on markers of the phase, x=0 on the other markers, and
the other coefficients are looked for). Each pixel is attributed the label
for which it has a maximal value of x. The Laplacian L of the image
is defined as:
- L_ii = d_i, the number of neighbors of pixel i (the degree of i)
- L_ij = -w_ij if i and j are adjacent pixels
The weight w_ij is a decreasing function of the norm of the local gradient.
This ensures that diffusion is easier between pixels of similar values.
When the Laplacian is decomposed into blocks of marked and unmarked
pixels::
L = M B.T
B A
with first indices corresponding to marked pixels, and then to unmarked
pixels, minimizing x.T L x for one phase amount to solving::
A x = - B x_m
where x_m = 1 on markers of the given phase, and 0 on other markers.
This linear system is solved in the algorithm using a direct method for
small images, and an iterative method for larger images.
Examples
--------
>>> np.random.seed(0)
>>> a = np.zeros((10, 10)) + 0.2 * np.random.rand(10, 10)
>>> a[5:8, 5:8] += 1
>>> b = np.zeros_like(a)
>>> b[3, 3] = 1 # Marker for first phase
>>> b[6, 6] = 2 # Marker for second phase
>>> random_walker(a, b)
array([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 2, 2, 2, 1, 1],
[1, 1, 1, 1, 1, 2, 2, 2, 1, 1],
[1, 1, 1, 1, 1, 2, 2, 2, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1, 1]], dtype=int32)
"""
# Parse input data
if mode is None:
if amg_loaded:
mode = 'cg_mg'
elif UmfpackContext is not None:
mode = 'cg'
else:
mode = 'bf'
if (labels != 0).all():
warn('Random walker only segments unlabeled areas, where '
'labels == 0. No zero valued areas in labels were '
'found. Returning provided labels.')
if return_full_prob:
# Find and iterate over valid labels
unique_labels = np.unique(labels)
unique_labels = unique_labels[unique_labels > 0]
out_labels = np.empty(labels.shape + (len(unique_labels), ),
dtype=np.bool)
for n, i in enumerate(unique_labels):
out_labels[..., n] = (labels == i)
else:
out_labels = labels
return out_labels
if data.ndim < 4:
raise ValueError('Dta must have 4 ' 'dimensions.')
dims = data[..., 0].shape # To reshape final labeled result
data = img_as_float(data)
# Spacing kwarg checks
if spacing is None:
spacing = np.asarray((1., ) * 4)
elif len(spacing) == len(dims):
if len(spacing) == 2: # Need a dummy spacing for singleton 3rd dim
spacing = np.r_[spacing, 1.]
else: # Convert to array
spacing = np.asarray(spacing)
else:
raise ValueError('Input argument `spacing` incorrect, should be an '
'iterable with one number per spatial dimension.')
if copy:
labels = np.copy(labels)
label_values = np.unique(labels)
# Reorder label values to have consecutive integers (no gaps)
if np.any(np.diff(label_values) != 1):
mask = labels >= 0
labels[mask] = rank_order(labels[mask])[0].astype(labels.dtype)
labels = labels.astype(np.int32)
# If the array has pruned zones, be sure that no isolated pixels
# exist between pruned zones (they could not be determined)
if np.any(labels < 0):
filled = ndi.binary_propagation(labels > 0, mask=labels >= 0)
labels[np.logical_and(np.logical_not(filled), labels == 0)] = -1
del filled
labels = np.atleast_3d(labels)
if np.any(labels < 0):
lap_sparse = _build_laplacian(data,
spacing,
mask=labels >= 0,
alpha=alpha,
beta=beta,
gamma=gamma,
a=a,
b=b,
c=c)
else:
lap_sparse = _build_laplacian(data,
spacing,
alpha=alpha,
beta=beta,
gamma=gamma,
a=a,
b=b,
c=c)
lap_sparse, B = _buildAB(lap_sparse, labels)
# We solve the linear system
# lap_sparse X = B
# where X[i, j] is the probability that a marker of label i arrives
# first at pixel j by anisotropic diffusion.
if mode == 'cg':
X = _solve_cg(lap_sparse,
B,
tol=tol,
return_full_prob=return_full_prob)
if mode == 'cg_mg':
if not amg_loaded:
warn("""pyamg (http://pyamg.org/)) is needed to use
this mode, but is not installed. The 'cg' mode will be used
instead.""")
X = _solve_cg(lap_sparse,
B,
tol=tol,
return_full_prob=return_full_prob)
else:
X = _solve_cg_mg(lap_sparse,
B,
tol=tol,
return_full_prob=return_full_prob)
if mode == 'bf':
X = _solve_bf(lap_sparse, B, return_full_prob=return_full_prob)
# Clean up results
if return_full_prob:
labels = labels.astype(np.float)
X = np.array([
_clean_labels_ar(Xline, labels, copy=True).reshape(dims)
for Xline in X
])
for i in range(1, int(labels.max()) + 1):
mask_i = np.squeeze(labels == i)
X[:, mask_i] = 0
X[i - 1, mask_i] = 1
else:
X = _clean_labels_ar(X + 1, labels).reshape(dims)
return X
# construct the argument parser and parse the arguments
#ap = argparse.ArgumentParser()
#ap.add_argument("-i", "--image", required = True, help = "Path to the image")
#args = vars(ap.parse_args())
train = []
# load the image and apply SLIC and extract (approximately)
# the supplied number of segments
#image = cv2.imread(args["image"])
for i in range(3, 20):
image = cv2.imread('images/image' + str(i) + '.jpg')
#img = image.load_img('image2.jpg')
ground_truth = cv2.imread('ground_truth/ground_truth' + str(i) +
'.jpg')
segments = slic(img_as_float(image), n_segments=100, sigma=5)
res = np.zeros(image.shape[:2], dtype="uint8")
selected = np.zeros(image.shape[:2], dtype="uint8")
res = ground_truth[:, :, 0]
# show the output of SLIC
fig = plt.figure("Superpixels")
ax = fig.add_subplot(1, 1, 1)
ax.imshow(
mark_boundaries(
img_as_float(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)),
segments))
plt.axis("off")
plt.show()
label = np.zeros(1)
if np.issubdtype(dtype, np.integer):
assert_array_equal(bin_centers, np.arange(imin, imax + 1))
assert frequencies[0][0] == channel_size
assert frequencies[0][-1] == 0
assert frequencies[1][0] == 0
assert frequencies[1][-1] == channel_size
# Test histogram equalization
# ===========================
np.random.seed(0)
test_img_int = data.camera()
# squeeze image intensities to lower image contrast
test_img = util.img_as_float(test_img_int)
test_img = exposure.rescale_intensity(test_img / 5. + 100)
def test_equalize_uint8_approx():
"""Check integer bins used for uint8 images."""
img_eq0 = exposure.equalize_hist(test_img_int)
img_eq1 = exposure.equalize_hist(test_img_int, nbins=3)
np.testing.assert_allclose(img_eq0, img_eq1)
def test_equalize_ubyte():
img = util.img_as_ubyte(test_img)
img_eq = exposure.equalize_hist(img)
cdf, bin_edges = exposure.cumulative_distribution(img_eq)
def validate_mask(val_loader, model, criterion, val_img_index):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to evaluate mode
model.eval()
end = time.time()
count = 0
for i, (input, target) in enumerate(val_loader):
count += 1
if count > val_img_index:
break
target = target.cuda(async=True)
input_var = torch.autograd.Variable(input, volatile=True).cuda()
target_var = torch.autograd.Variable(target, volatile=True).cuda()
img_show = input[0].numpy().copy()
img_show = img_show.transpose(1, 2, 0)
img_show -= img_show.min()
img_show /= img_show.max()
img_show *= 255
img_show = img_show.astype(np.uint8)
if count == val_img_index:
#cv2.imwrite("orginal_img.png", img_show)
segments = felzenszwalb(img_as_float(img_show),
scale=100,
sigma=0.5,
min_size=50)
print("Felzenszwalb number of segments: {}".format(
len(np.unique(segments))))
# cv2.imshow('superpixels', mark_boundaries(img_as_float(img_show), segments))
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# compute output
output = model(input_var)
loss = criterion(output, target_var)
pred = output.data.max(1, keepdim=True)[1]
label = target[0]
if pred[0].cpu().numpy()[0] == label:
print("correct prediction, index ", count)
correct_pred_count = 0
wrong_pred_count = 0
dict_pixel = get_pixel_sorted_mask_label()
plot_summed_heatmap(val_img_index, img_show, label, dict_pixel)
sorted_dict_values_set = sorted(set(dict_pixel.values()))
# Binary search the pixel label threshold
first = 0
last = len(sorted_dict_values_set) - 1
while first <= last:
midpoint = int((first + last) / 2)
mask_threshold1 = sorted_dict_values_set[midpoint]
mask_threshold2 = sorted_dict_values_set[midpoint + 1]
mask1 = generate_new_mask(dict_pixel, mask_threshold1)
mask2 = generate_new_mask(dict_pixel, mask_threshold2)
print("sorted_dict_values_set")
print(sorted_dict_values_set)
print("len(sorted_dict_values_set)")
print(len(sorted_dict_values_set))
masked_img1 = input[0].numpy().copy() * mask1
masked_img2 = input[0].numpy().copy() * mask2
masked_img_batch1 = masked_img1[None, :, :, :]
masked_img_batch2 = masked_img2[None, :, :, :]
masked_img_tensor1 = torch.autograd.Variable(
torch.from_numpy(masked_img_batch1)).cuda()
masked_img_tensor2 = torch.autograd.Variable(
torch.from_numpy(masked_img_batch2)).cuda()
# Evaluate the NN model
mask_output1 = model(masked_img_tensor1)
mask_output2 = model(masked_img_tensor2)
pred_mask1 = mask_output1.data.max(1, keepdim=True)[1]
pred_mask2 = mask_output2.data.max(1, keepdim=True)[1]
masked_img_show1 = masked_img1.copy()
masked_img_show1 = masked_img_show1.transpose(1, 2, 0)
masked_img_show1 -= masked_img_show1.min()
masked_img_show1 /= masked_img_show1.max()
masked_img_show1 *= 255
masked_img_show1 = masked_img_show1.astype(np.uint8)
masked_img_show2 = masked_img2.copy()
masked_img_show2 = masked_img_show2.transpose(1, 2, 0)
masked_img_show2 -= masked_img_show2.min()
masked_img_show2 /= masked_img_show2.max()
masked_img_show2 *= 255
masked_img_show2 = masked_img_show2.astype(np.uint8)
if pred_mask1[0].cpu().numpy()[0] == target[0]:
correct_pred_count += 1
print("correct_pred_count: ", correct_pred_count)
if pred_mask2[0].cpu().numpy()[0] != target[0]:
plt.subplot(141), plt.imshow(
cv2.cvtColor(img_show, cv2.COLOR_BGR2RGB),
'gray'), plt.title('original_img_label_{}'.format(
classes_dict[target[0]]),
fontsize=60)
plt.subplot(142), plt.imshow(
mark_boundaries(img_as_float(img_show[:, :, ::-1]),
segments),
'gray'), plt.title('Superpixel', fontsize=60)
plt.subplot(143), plt.imshow(
mask1 * 255, 'gray'), plt.title(
"Mask threshold_{}".format(mask_threshold1),
fontsize=60)
plt.subplot(144), plt.imshow(
cv2.cvtColor(masked_img_show1, cv2.COLOR_BGR2RGB),
'gray'), plt.title(
'Org_img_with_mask pred_{}'.format(
classes_dict[pred_mask1[0].cpu().numpy()
[0]]),
fontsize=60)
#cv2.imwrite("frog.png", masked_img_show1)
figure = plt.gcf() # get current figure
figure.set_size_inches(90, 30)
plt.savefig(
'result_imgs/index_{}_threshold_{}.png'.format(
val_img_index, mask_threshold1))
plt.subplot(141), plt.imshow(
cv2.cvtColor(img_show, cv2.COLOR_BGR2RGB),
'gray'), plt.title('original_img_label_{}'.format(
classes_dict[target[0]]),
fontsize=60)
plt.subplot(142), plt.imshow(
mark_boundaries(img_as_float(img_show[:, :, ::-1]),
segments),
'gray'), plt.title('Superpixel', fontsize=60)
plt.subplot(143), plt.imshow(
mask2 * 255, 'gray'), plt.title(
"Mask threshold_{}".format(mask_threshold2),
fontsize=60)
plt.subplot(144), plt.imshow(
cv2.cvtColor(masked_img_show2, cv2.COLOR_BGR2RGB),
'gray'), plt.title('pred_{}'.format(
classes_dict[pred_mask2[0].cpu().numpy()[0]]),
fontsize=60)
figure = plt.gcf() # get current figure
figure.set_size_inches(90, 30)
plt.savefig(
'result_imgs/index_{}_threshold_{}.png'.format(
val_img_index, mask_threshold2))
# plt.show()
# plt.close()
return mask_threshold1
else:
first = midpoint + 1
else:
wrong_pred_count += 1
print("wrong_pred_count: ", wrong_pred_count)
last = midpoint - 1
print("masked label threshold")
print(mask_threshold1)
def imgs2lf(input_dir,
lf_file,
lf_dataset='lightfield',
img_extension='.png',
dtype=np.uint8,
RGB=True):
"""
Convert several images to a lightfield.
Parameters
----------
input_dir : string
The directory where the ligthfield images are located.
lf_file: string
The filename (including the directory), of the output file.
lf_dataset : string, optional
The new container name of the hdf5 file.
img_extension : string, optional
The file extension of the images to look for.
dtype : numpy.dtype, optional
The new data type for the downscaled lightfield. Must be either
np.float64, np.uint8 or np.uint16.
RGB : bool, optional
If True, the output lightfield will be converted to RGB (default).
Otherwise gray type images are stored.
"""
# look for images
files = multiLoading(identifier="*.{e}".format(e=img_extension),
path=input_dir)
# prepare saving
lf_file = prepareSaving(lf_file, extension=".hdf5")
# Which dtype should be used?
if dtype == np.float64:
img_0 = img_as_float(imread(files[0]))
elif dtype == np.uint8:
img_0 = img_as_ubyte(imread(files[0]))
elif dtype == np.uint16:
img_0 = img_as_uint(imread(files[0]))
else:
raise TypeError('The given data type is not supported!')
# Do we shall take RGB or gray images?
if (len(img_0.shape) == 3 and img_0.shape[2] == 3):
rows, cols, orig_channels = img_0.shape # automatically determine the images'shapes from the first image.
elif len(img_0.shape) == 2:
orig_channels = 1
rows, cols = img_0.shape # automatically determine the images'shapes from the first image.
else:
raise TypeError('The given images are neither gray nor RGB images!')
f_out = h5py.File(lf_file, 'w')
if RGB:
dataset = f_out.create_dataset(lf_dataset,
shape=(len(files), rows, cols, 3),
dtype=dtype)
else:
dataset = f_out.create_dataset(lf_dataset,
shape=(len(files), rows, cols),
dtype=dtype)
for k in range(len(files)):
if dtype == np.float64:
if orig_channels == 1 and RGB:
dataset[k, ...] = img_as_float(gray2rgb(imread(files[k])))
elif orig_channels == 3 and not RGB:
dataset[k, ...] = img_as_float(rgb2gray(imread(files[k])))
else:
dataset[k, ...] = img_as_float(imread(files[k]))
elif dtype == np.uint8:
if orig_channels == 1 and RGB:
dataset[k, ...] = img_as_ubyte(gray2rgb(imread(files[k])))
elif orig_channels == 3 and not RGB:
dataset[k, ...] = img_as_ubyte(rgb2gray(imread(files[k])))
else:
dataset[k, ...] = img_as_ubyte(imread(files[k]))
elif dtype == np.uint16:
if orig_channels == 1 and RGB:
dataset[k, ...] = img_as_uint(gray2rgb(imread(files[k])))
elif orig_channels == 3 and not RGB:
dataset[k, ...] = img_as_uint(rgb2gray(imread(files[k])))
else:
dataset[k, ...] = img_as_uint(imread(files[k]))
else:
raise TypeError('Given dtype not supported.')
f_out.close()
def test_otsu_coins_image_as_float():
coins = util.img_as_float(data.coins())
assert 0.41 < threshold_otsu(coins) < 0.42
#####################################################################
# Calibrating a wavelet denoiser
import numpy as np
from matplotlib import pyplot as plt
from skimage.data import chelsea
from skimage.restoration import calibrate_denoiser, denoise_wavelet
from skimage.util import img_as_float, random_noise
from functools import partial
# rescale_sigma=True required to silence deprecation warnings
_denoise_wavelet = partial(denoise_wavelet, rescale_sigma=True)
image = img_as_float(chelsea())
sigma = 0.3
noisy = random_noise(image, var=sigma**2)
# Parameters to test when calibrating the denoising algorithm
parameter_ranges = {
'sigma': np.arange(0.1, 0.3, 0.02),
'wavelet': ['db1', 'db2'],
'convert2ycbcr': [True, False],
'multichannel': [True]
}
# Denoised image using default parameters of `denoise_wavelet`
default_output = denoise_wavelet(noisy, multichannel=True, rescale_sigma=True)
# Calibrate denoiser
def validate(val_loader, model, criterion, eval_img_index):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to evaluate mode
model.eval()
count = 0
for i, (input, target) in enumerate(val_loader):
count += 1
if count > eval_img_index:
break
input_var = torch.autograd.Variable(input, volatile=True).cuda()
target_var = torch.autograd.Variable(target, volatile=True).cuda()
img_show = input[0].numpy().copy()
img_show = img_show.transpose(1, 2, 0)
img_show -= img_show.min()
img_show /= img_show.max()
img_show *= 255
img_show = img_show.astype(np.uint8)
# cv2.imshow('index_{}_label_{}'.format(i, classes_dict[target[0]]), img_show)
# # cv2.waitKey(0)
if count == eval_img_index:
segments = felzenszwalb(img_as_float(img_show),
scale=100,
sigma=0.5,
min_size=50)
print("Felzenszwalb number of segments: {}".format(
len(np.unique(segments))))
#cv2.imshow('superpixels', mark_boundaries(img_as_float(img_show), segments))
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# compute output
output = model(input_var)
loss = criterion(output, target_var)
print("output")
print(output)
print("target_var")
print(target_var)
print("loss")
print(loss)
pred = output.data.max(1, keepdim=True)[1]
label = target[0]
print("label ", label)
print("pred[0].cpu().numpy() ", pred[0].cpu().numpy()[0])
mask_dir = './masks'
if not os.path.exists(mask_dir):
os.makedirs(mask_dir)
else:
shutil.rmtree(mask_dir) #removes all the subdirectories!
os.makedirs(mask_dir)
if pred[0].cpu().numpy()[0] == label:
print("correct prediction, index_{} , label_{}".format(
count, classes_dict[label]))
correct_pred_count = 0
wrong_pred_count = 0
for i in range(args.num_mask_samples):
total_num_segments = len(np.unique(segments))
num_conse_superpixels = int(0.4 * total_num_segments)
print("total_num_segments: ", total_num_segments)
print("num_conse_superpixels: ", num_conse_superpixels)
firstIndex = randint(
1, total_num_segments - num_conse_superpixels)
random_sampled_list = np.unique(segments)[firstIndex:(
firstIndex + num_conse_superpixels)]
#random_sampled_list= sample(range(np.unique(segments)[0], np.unique(segments)[-1]), num_conse_superpixels)
#print("random_sampled_list: ", random_sampled_list)
mask = np.zeros(img_show.shape[:2], dtype="uint8")
#mask.fill(1)
for (j, segVal) in enumerate(random_sampled_list):
mask[segments == segVal] = 1
masked_img = input[0].numpy().copy() * mask
masked_img_batch = masked_img[None, :, :, :]
masked_img_tensor = torch.autograd.Variable(
torch.from_numpy(masked_img_batch)).cuda()
mask_output = model(masked_img_tensor)
pred_mask = mask_output.data.max(1, keepdim=True)[1]
masked_img_show = masked_img.copy()
masked_img_show = masked_img_show.transpose(1, 2, 0)
masked_img_show -= masked_img_show.min()
masked_img_show /= masked_img_show.max()
masked_img_show *= 255
masked_img_show = masked_img_show.astype(np.uint8)
if pred_mask[0].cpu().numpy()[0] == target[0]:
correct_pred_count += 1
print("correct_pred_count: ", correct_pred_count)
cv2.imwrite('./masks/mask_{}_{}.png'.format(i, 1),
mask * 255)
# cv2.imwrite('./mask_on_img/masked_imgs_{}.png'.format(i), masked_img_show)
else:
wrong_pred_count += 1
print("wrong_pred_count: ", wrong_pred_count)
cv2.imwrite('./masks/mask_{}_{}.png'.format(i, 0),
mask * 255)
#cv2.imwrite('./mask_on_img/masked_imgs_{}.png'.format(i), masked_img_show)
return correct_pred_count
else:
print("wrong prediction")
#print("%d samples, the corrrect prediction number: %d "%(len(mask_filenames), correct_pred_count))
return 0
#total_int = dapi.sum( axis = 0 )
global_thresh = filters.threshold_otsu(nuclei) * 5
mask_nuclei = nuclei > global_thresh
#mask_nuclei = filters.threshold_local(nuclei, 21)
mask_roi = np.zeros(nuclei.shape, dtype=bool)
#mask_roi[nuclei>filters.threshold_otsu(nuclei)] = True
mask_roi[nuclei > 1000] = True
roi_ilot = morphology.remove_small_objects(mask_roi, 500000)
bbox = ndi.find_objects(roi_ilot)
dapi_roi = dapi[bbox[0]]
#rb_roi = rb[bbox[0]]
# Denoise using chambolle
dapi_float = util.img_as_float(dapi_roi)
dapi_float_tvc = restoration.denoise_tv_chambolle(dapi_float, weight=0.05)
# Find z-stack w/highest variance
variance = [np.var(image) for _, image in enumerate(mask_dapi)]
idx = np.argmax(variance)
dapi_roi_z = dapi_float_tvc[idx - 3:idx + 3]
distance_a = ndi.distance_transform_edt(mask_dapi[3])
#Local peak markers
smooth_distance = filters.gaussian(distance_a, sigma=4)
local_maxi = feature.peak_local_max(smooth_distance,
indices=True,
exclude_border=False,
footprint=np.ones((15, 15)))
markers_lm = np.zeros(distance_a.shape, dtype=np.int)
for i, h in enumerate(norm_hists):
if i in fgbg_superpixels[0]:
g.add_tedge(nodes[i], 0, 1000) # FG - set high cost to BG
elif i in fgbg_superpixels[1]:
g.add_tedge(nodes[i], 1000, 0) # BG - set high cost to FG
else:
g.add_tedge(nodes[i],
cv2.compareHist(fgbg_hists[0], h, hist_comp_alg),
cv2.compareHist(fgbg_hists[1], h, hist_comp_alg))
g.maxflow()
return g.get_grid_segments(nodes)
if __name__ == '__main__':
img = img_as_float(astronaut()[::2, ::2])
img_marking = cv2.imread("astronaut_marking.png")
centers, colors_hists, segments, neighbors = superpixels_histograms_neighbors(
img)
fg_segments, bg_segments = find_superpixels_under_marking(
img_marking, segments)
# get cumulative BG/FG histograms, before normalization
fg_cumulative_hist = cumulative_histogram_for_superpixels(
fg_segments, colors_hists)
bg_cumulative_hist = cumulative_histogram_for_superpixels(
bg_segments, colors_hists)
norm_hists = normalize_histograms(colors_hists)
def load_pred(output_dir, imname):
pred_path = os.path.join(output_dir, '{}.png'.format(imname))
return img_as_float(imread(pred_path))
# prepare filter bank kernels
kernels = []
for theta in range(4):
theta = theta / 4. * np.pi
for sigma in (1, 3):
for frequency in (0.05, 0.25):
kernel = np.real(
gabor_kernel(frequency,
theta=theta,
sigma_x=sigma,
sigma_y=sigma))
kernels.append(kernel)
shrink = (slice(0, None, 3), slice(0, None, 3))
brick = img_as_float(data.brick())[shrink]
grass = img_as_float(data.grass())[shrink]
gravel = img_as_float(data.gravel())[shrink]
image_names = ('brick', 'grass', 'gravel')
images = (brick, grass, gravel)
# prepare reference features
ref_feats = np.zeros((3, len(kernels), 2), dtype=np.double)
ref_feats[0, :, :] = compute_feats(brick, kernels)
ref_feats[1, :, :] = compute_feats(grass, kernels)
ref_feats[2, :, :] = compute_feats(gravel, kernels)
print('Rotated images matched against references using Gabor filter banks:')
print('original: brick, rotated: 30deg, match result: ', end='')
feats = compute_feats(ndi.rotate(brick, angle=190, reshape=False), kernels)
from __future__ import print_function
import numpy as np
from matplotlib import pyplot as plt
from skimage import data
from skimage.util import img_as_float
from skimage.feature import (corner_harris, corner_subpix, corner_peaks,
plot_matches)
from skimage.transform import warp, AffineTransform
from skimage.exposure import rescale_intensity
from skimage.color import rgb2gray
from skimage.measure import ransac
# generate synthetic checkerboard image and add gradient for the later matching
checkerboard = img_as_float(data.checkerboard())
img_orig = np.zeros(list(checkerboard.shape) + [3])
img_orig[..., 0] = checkerboard
gradient_r, gradient_c = np.mgrid[0:img_orig.shape[0],
0:img_orig.shape[1]] / float(
img_orig.shape[0])
img_orig[..., 1] = gradient_r
img_orig[..., 2] = gradient_c
img_orig = rescale_intensity(img_orig)
img_orig_gray = rgb2gray(img_orig)
# warp synthetic image
tform = AffineTransform(scale=(0.9, 0.9), rotation=0.2, translation=(20, -10))
img_warped = warp(img_orig, tform.inverse, output_shape=(200, 200))
img_warped_gray = rgb2gray(img_warped)
def paste(img, canvas, i, j, method='replace', export_dtype='float'):
""" paste `img` on `canvas` with its left-top corner at (i, j) """
# check dtypes
img = su.img_as_float(img)
canvas = su.img_as_float(canvas)
# check shapes
if len(img.shape) != 2 or len(img.shape) != 3:
if len(canvas.shape) != 2 or len(canvas.shape) != 3:
assert AttributeError(
'dimensions of input images not all equal to 2 or 3!')
# check channels
# all grayscale image
if len(img.shape) == 2 and len(canvas.shape) == 2:
pass
# `img` color image, possible with alpha channel; `canvas` grayscale image
elif len(img.shape) == 3 and len(canvas.shape) == 2:
c = img.shape[-1]
if c == 3:
canvas = np.stack([canvas] * c, axis=-1)
if c == 4:
canvas = np.stack([canvas] * (c - 1) +
[np.ones((canvas.shape[0], canvas.shape[1]))],
axis=-1)
# `canvas` color image, possible with alpha channel; `img` grayscale image
elif len(img.shape) == 2 and len(canvas.shape) == 3:
c = canvas.shape[-1]
if c == 3:
img = np.stack([img] * c, axis=-1)
if c == 4:
img = np.stack([img] * (c - 1) +
[np.ones((img.shape[0], img.shape[1]))],
axis=-1)
# all color image
elif len(img.shape) == 3 and len(canvas.shape) == 3:
if img.shape[-1] == 3 and canvas.shape[-1] == 4:
img = np.concatenate(
[img, np.ones((img.shape[0], img.shape[1], 1))], -1)
elif img.shape[-1] == 4 and canvas.shape[-1] == 3:
canvas = np.concatenate(
[canvas,
np.ones((canvas.shape[0], canvas.shape[1], 1))], -1)
elif img.shape[-1] == canvas.shape[-1]:
pass
else:
assert ValueError('channel number should equal to 3 or 4!')
# get shapes
h_i, w_i = img.shape[:2]
h_c, w_c = canvas.shape[:2]
# find extent of `img` on `canvas`
i_min = np.max([0, i])
i_max = np.min([h_c, i + h_i])
j_min = np.max([0, j])
j_max = np.min([w_c, j + w_i])
# paste `img` on `canvas`
if method == 'replace':
canvas[i_min:i_max, j_min:j_max] = img[i_min - i:i_max - i,
j_min - j:j_max - j]
elif method == 'add':
canvas[i_min:i_max, j_min:j_max] += img[i_min - i:i_max - i,
j_min - j:j_max - j]
else:
raise ValueError('no such method!')
# return `canvas`
if export_dtype == 'float':
return canvas
elif export_dtype == 'ubyte':
return su.img_as_ubyte(canvas)
else:
raise ValueError('no such data type for exporting!')
def test_yen_coins_image_as_float():
coins = util.img_as_float(data.coins())
assert 0.43 < threshold_yen(coins) < 0.44
bck = bck / i
#kp.append(list(bck.astype(np.uint8)))
return list(bck.astype(np.uint8))
# def read_raw(filename,r,c):
# fd = open(os.path.join(mask_src,filename.split('.')[0] + '.raw' ), 'rb')
# f = np.fromfile(fd, dtype=np.uint8,count=r*c)
# im = f.reshape((r,c)) #notice row, column format
# fd.close()
# return im
g = gm.GenerateMask()
for filename in file_list:
# load the image and convert it to a floating point data type
image = img_as_float(io.imread(os.path.join(img_src, filename)))
img = cv2.imread(os.path.join(img_src, filename))
#image = cv2.resize(image,None,fx=2.5,fy=2.5,interpolation=cv2.INTER_CUBIC)
#image = cv2.resize(image,None,fx=2.5,fy=2.5,interpolation=cv2.INTER_CUBIC)
#img = cv2.resize(img,None,fx=2.5,fy=2.5,interpolation=cv2.INTER_CUBIC).astype(np.uint8)
temp = image.copy().astype('uint8')
h, w, d = image.shape
kp = read_csv(filename)
bg, sure_fg = g.mask_generate(kp, h, w)
mid_hip = ret_mid_pt([kp[8], kp[11]])
sure_fg = add_limb(kp[1], mid_hip, sure_fg)
# sure_fg2 = add_limb(kp[1],kp[11],sure_fg)
# sure_fg2 = add_limb(kp[1],kp[8],sure_fg2)
sure_fg = add_limb(kp[5], kp[8], sure_fg)