def func(frame):
_dtype = frame.dtype
kernel = mor.disk(3)
frameWP = frame - mor.white_tophat(frame, kernel) * (mor.white_tophat(frame, kernel) > 1000).astype(float)
kernel = mor.rectangle(25, 1)
closed = mor.closing(frameWP, kernel)
opened = mor.opening(closed, kernel)
result = ((frameWP.astype(float) / opened.astype(float)) * 3000.0)
return result.astype(_dtype)
def white_top_func(filename):
image = Image.open(filename)
image = asarray(image)
image = rgb2gray(image)
image = morphology.white_tophat(image, disk(6))
plt.imsave(filename, image, cmap='gray')
return filename
def segment_mask(img_GRAY):
'''
Computation of Segmentation Mask for grayscale images
mask_pred = segment_mask(img_GRAY)
INPUTS:
img_GRAY ~ 2D array, dtype=float (N x M) [Grayscale input for which mask will be generated
OUTPUTS:
mask_pred ~ Boolean array (N x M) [Output segmentation mask in [0,1]]
'''
# Increase contrast by adapative histogram equalization
# The low contrast in large parts of the image, coupled with the high
# intensity for the bottom-right artifacts, motivated selection of a locally
# adaptive form of histogram equalization.
img_HISTEQ = exposure.equalize_adapthist(img_GRAY)
# Use triangle threshold for segmenting objects
# Given the skewed and unimodal distribution of pixel values,
# the triangle threshold was selected as desirable for isolating the
# tail of peak pixel intensities without including brighter segments of the
# tissue (as was observed with Otsu)
thresh = filters.threshold_triangle(img_HISTEQ)
mask_init = img_HISTEQ > thresh
# Use white tophat filtering to remove small artifacts
# White top-hat filtering inflates objects larger than the structural element,
# and then returns the complement. The original mask can be used to crush
# small structures in the original image.
strel = morphology.disk(2)
noise = morphology.white_tophat(mask_init, strel)
mask_pred = np.logical_and(mask_init, np.logical_not(noise))
return mask_pred
def subtract_background(image, elem='disk', radius=50, light_bg=False):
"""Background substraction using structure element.
Slightly adapted from: https://forum.image.sc/t/background-subtraction-in-scikit-image/39118/4
:param image: input image
:type image: NumPy.Array
:param elem: type of the structure element, defaults to 'disk'
:type elem: str, optional
:param radius: size of structure element [pixel], defaults to 50
:type radius: int, optional
:param light_bg: light background, defaults to False
:type light_bg: bool, optional
:return: image with background subtracted
:rtype: NumPy.Array
"""
# use 'ball' here to get a slightly smoother result at the cost of increased computing time
if elem == 'disk':
str_el = disk(radius)
if elem == 'ball':
str_el = ball(radius)
if light_bg:
img_subtracted = black_tophat(image, str_el)
if not light_bg:
img_subtracted = white_tophat(image, str_el)
return img_subtracted
def _white_tophat(self, image: Union[xr.DataArray,
np.ndarray]) -> np.ndarray:
if self.is_volume:
structuring_element = ball(self.masking_radius)
else:
structuring_element = disk(self.masking_radius)
return white_tophat(image, selem=structuring_element)
def white_hatting(file='zones/0.png', use_t=False, t='otsu', rad=3):
from skimage.morphology import white_tophat
from skimage.morphology import disk
if use_t:
img = threshold(file, t=t)
else:
img = misc.imread(file)
selem = disk(rad)
wh = white_tophat(img, selem)
fig, (ax1, ax2) = plt.subplots(ncols=2,
figsize=(8, 4),
sharex=True,
sharey=True)
ax1.imshow(img, cmap=plt.cm.gray)
ax1.set_title('original')
ax1.axis('off')
ax1.set_adjustable('box-forced')
ax2.imshow(wh, cmap=plt.cm.gray)
ax2.set_title('white')
ax2.axis('off')
ax2.set_adjustable('box-forced')
plt.show()
def getImage(i, imOriginal, mean):
if i == 0: #return a central crop of 99x99 pixels
resized = cv2.resize(imOriginal, (128, 128))
resized = resized * 256.0 - mean
return resized[14:113, 14:113]
rot = np.random.uniform(0, 360, 1).astype(int)[0] #Random rotations
rot = 90 * np.random.uniform(0, 4, 1).astype(int)[0] #Random rotations
# rot = 0
im_size = imOriginal.shape[0]
if (np.random.rand() > 0.5):
if (np.random.rand() > 0.5):
imOriginal = cv2.flip(imOriginal, 0)
else:
imOriginal = cv2.flip(imOriginal, 1)
scale = np.random.uniform(0.9, 1.1)
mat = cv2.getRotationMatrix2D((im_size / 2, im_size / 2), rot, scale=scale)
resized = cv2.warpAffine(img_out,
mat, (im_size, im_size),
borderValue=(255, 255, 255))
img_out = np.zeros((resized.shape[0], resized.shape[1], 3), dtype=np.uint8)
img_orig = resized[:, :, 0]
img_btop = 255 - black_tophat(img_orig, selem)
img_wtop = 255 - white_tophat(img_orig, selem)
img_out[:, :, 1] = img_btop
img_out[:, :, 2] = img_wtop
resized = cv2.resize(img_out, (128, 128))
resized = resized * 256.0 - mean #Geht richtig in den Keller auf Werte um 4, wenn man mean nicht abzieht
# offsetX = np.random.uniform(10,18,1).astype(int)[0]
# offsetY = np.random.uniform(10,18,1).astype(int)[0]
offsetX = np.random.uniform(0, 28, 1).astype(int)[0] #Random rotations
offsetY = np.random.uniform(0, 28, 1).astype(int)[0] #Random rotations
return resized[offsetY:(99 + offsetY), offsetX:(99 + offsetX)]
def generate_cost_image(image):
"""
Convert an image to be used for A* route finding
Generate an image where each pixel is a cost rather than an intensity. Since
the path finding algorithms are designed to ascend gradients/follow high
intensity paths, we need to invert the image before using it.
"""
disk_size = 5
original_scale = 10.0
cost_image = image.copy()
selem = disk(disk_size)
image_processed = white_tophat(image, selem)
# Cutoff anything over the max of the top-hatted image
cost_image[cost_image > numpy.max(image_processed)] = numpy.max(
image_processed)
# Create a cost image scaled to 10% of the processed image
cost_image = (cost_image / numpy.max(image_processed)) * (
numpy.max(image_processed) / original_scale)
# Add the cost image and the processed image together
cost_image = cost_image + image_processed
# Invert
cost_image = (numpy.max(cost_image) - cost_image)
return cost_image
def bscan_cut(bscans, onh_ybox=[-1, -1]):
if onh_ybox != [-1, -1]:
#either no coordinates were entered, or they weren't detected
cut_scans = bscans[onh_ybox[0]:onh_ybox[1], :, :]
else:
cut_scans = bscans[120:136, :, :]
avg_scans = np.mean(cut_scans, axis=0)
#Email from Mayank to Eric on 8-15-2015 on Mayank and Robert's formula to resize bscans to 1 px**2
#height = y_height*0.84
#width = 1600 px
height = np.round(avg_scans.shape[1] * 0.84)
width = 1600
scan = tf.resize(avg_scans, (height, width), order=3, mode="reflect")
#drop to 8 bit, otherwise the averaging takes a very long time. Which seems strange...
scan_s = scan.astype("uint8")
scan_m = sf.rank.mean(scan_s, selem=mp.disk(50))
#Worried there are issues with top hat. May smear in noise that becomes part of the center of mass computation.
scan_b = mp.white_tophat(scan_m, selem=mp.square(500))
m = sm.moments(scan_b, order=1)
y = int(np.round(m[0, 1] / m[0, 0]))
ymin = y - 300
ymax = y + 300
cut_scan = scan[ymin:ymax, :]
return cut_scan.astype("float32")
def white_tophat():
global filterimage
kernel = morphology.disk(5)
img_white = morphology.white_tophat(img, kernel)
filterimage = img_white
io.imshow(img_white)
io.show()
def __init__(self, file, csv_file, nuclei_channel=1):
self.raw_im = np.squeeze(czi.imread(file))[nuclei_channel]
self.labeled_spots_im = np.squeeze(czi.imread(file))[0]
self.mask_1 = cv2.threshold(
self.raw_im,
skimage.filters.threshold_otsu(self.raw_im) + 152, 1,
cv2.THRESH_BINARY)[1]
self.nuclei_image = skimage.exposure.equalize_hist(
morphology.white_tophat(
np.squeeze(czi.imread(file))[nuclei_channel], square(70)),
mask=self.mask_1)
self.corrected_ni = np.multiply(
np.divide(self.nuclei_image, np.amax(self.nuclei_image)), 255)
self.otsu = skimage.filters.threshold_otsu(self.corrected_ni)
self.dtype = str(self.nuclei_image.dtype)
self.nuclei_mask = cv2.threshold(self.corrected_ni, self.otsu, 1,
cv2.THRESH_BINARY)[1]
self.nuclei_mask_dense = cv2.threshold(self.corrected_ni,
self.otsu + 80, 1,
cv2.THRESH_BINARY)[1]
self.nuc_chan = nuclei_channel
self.input_shape = self.nuclei_image.shape
self.filename = file
self.csv_file = csv_file
self.spatial_df = pd.read_csv(csv_file)
print('csv: \n', self.csv_file)
print('image: \n', self.filename)
def ExtractCandidates(im_norm,h,radius,nbit):
"""extract signal candidates applying h_maxima transform
INPUTS:
im_norm=normalised image,
h=h_maxima threshold,
radius=structuring element radius,
nbit= encoding"""
# Normalized top_hat filtering
se=disk(radius)
im=white_tophat(im_norm,se)
#filtering local maxima
h_maxima=extrema.h_maxima(im, h,selem=diamond(1))
label_h_max=label(h_maxima,neighbors=4)
labels=pd.DataFrame(data={'labels':np.sort(label_h_max[np.where(label_h_max!=0)])})
dup=labels.index[labels.duplicated() == True].tolist() #find duplicates labels (=connected components)
#splitting connected regions to get only one local maxima
max_mask=np.zeros(im.shape)
max_mask[label_h_max!=0]=np.iinfo(nbit).max
for i in range (len(dup)):
r,c=np.where(label_h_max==labels.loc[dup[i],'labels']) #find coord of points having the same label
meanpoint_x=np.mean(c)
meanpoint_y=np.mean(r)
dist=[distance.euclidean([meanpoint_y,meanpoint_x],[r[j],c[j]]) for j in range(len(r))]
ind=dist.index(min(dist))
r,c=np.delete(r,ind),np.delete(c,ind) #delete values at ind position.
max_mask[r,c]=0 #set to 0 points != medoid coordinates
return max_mask
def getImage(i, imOriginal, mean):
if i == 0: #return a central crop of 99x99 pixels
resized = cv2.resize(imOriginal, (128, 128))
resized = resized * 256.0 - mean
return resized[14:113, 14:113]
rot = np.random.uniform(0,360,1).astype(int)[0] #Random rotations
rot = 90 * np.random.uniform(0,4,1).astype(int)[0] #Random rotations
# rot = 0
im_size = imOriginal.shape[0]
if (np.random.rand() > 0.5):
if (np.random.rand() > 0.5):
imOriginal = cv2.flip(imOriginal,0)
else:
imOriginal = cv2.flip(imOriginal,1)
scale = np.random.uniform(0.9,1.1)
mat = cv2.getRotationMatrix2D((im_size / 2, im_size / 2), rot, scale=scale)
resized = cv2.warpAffine(img_out, mat, (im_size, im_size), borderValue=(255,255,255))
img_out = np.zeros((resized.shape[0], resized.shape[1], 3), dtype=np.uint8)
img_orig = resized[:,:,0]
img_btop = 255-black_tophat(img_orig, selem)
img_wtop = 255-white_tophat(img_orig, selem)
img_out[:, :, 1] = img_btop
img_out[:, :, 2] = img_wtop
resized = cv2.resize(img_out, (128, 128))
resized = resized * 256.0 - mean #Geht richtig in den Keller auf Werte um 4, wenn man mean nicht abzieht
# offsetX = np.random.uniform(10,18,1).astype(int)[0]
# offsetY = np.random.uniform(10,18,1).astype(int)[0]
offsetX = np.random.uniform(0,28,1).astype(int)[0] #Random rotations
offsetY = np.random.uniform(0,28,1).astype(int)[0] #Random rotations
return resized[offsetY:(99+offsetY), offsetX:(99+offsetX)]
def adjust_image(image,
lower_thresh=2, upper_thresh=98,
filter_size=0,
cmap='jet'
):
"""
Applies contrast stretching to an image, then
uses a white top-hat filter to remove small patches
of brightness that would cause false positives later.
Input: image; values for min and max percentile
brightnesses to keep; size below which bright patches
will be removed; colourmap.
Output: image, hopefully with most of the background stripped
out. If it's worked well, it'll look like bright blobs on a
dark background.
"""
p2, p98 = np.percentile(image, (lower_thresh, upper_thresh))
img_rescale = exposure.rescale_intensity(image, in_range=(p2, p98))
selem = morphology.disk(filter_size)
wht_tophat = morphology.white_tophat(img_rescale,selem=selem)
io.imshow(img_rescale - wht_tophat, cmap=cmap)
return img_rescale
def detectCentroid(input):
# Centroid Detection and contouring
output = white_tophat(input, square(35))
blur = cv2.GaussianBlur(output, (5, 5), 0)
_, threshold = cv2.threshold(blur, 254, 255, 0)
threshold = threshold.astype('uint8')
contours, hierarchy = cv2.findContours(threshold, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
i = 0
for cnt in contours:
for i in tqdm(np.arange(len(contours))):
cv2.drawContours(threshold, [cnt], -1, (78, 55, -128), 1)
M = cv2.moments(cnt)
time.sleep(0.1)
i = i + 1
cX = int(M["m10"] / M["m00"])
cY = int(M["m01"] / M["m00"])
cv2.circle(threshold, (cX, cY), 2, (78, 55, -128), 1)
cv2.putText(threshold, "center", (cX - 20, cY - 20),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)
cv2.imshow("Centroid", threshold)
cv2.moveWindow("Centroid", 300, 30)
print("X-Coordinate: " + str(cX))
print("Y-Coordinate: " + str(cY))
cv2.destroyWindow("Original Image")
cv2.waitKey(0)
return threshold, cX, cY
def insert_db(self, mode, image, label, features, channel_no, inverse):
if inverse:
image_ubyte = 255 - img_as_ubyte(image)
else:
image_ubyte = img_as_ubyte(image)
image_ubyte = numpy.transpose(image_ubyte, (2, 0, 1))
image_string = image_ubyte.tostring()
if features != None:
delimeter = '!@#$'
self.datum.data = image_string + delimeter + features
elif channel_no > 3:
selem = disk(6)
w_tophat = white_tophat(image_ubyte, selem)
b_tophat = black_tophat(image_ubyte, selem)
self.datum.data = image_string + w_tophat.tostring() + b_tophat.tostring()
else:
self.datum.data = image_string
if label != None:
self.datum.label = int(label)
serialized = self.datum.SerializeToString()
if mode == 'train':
self.train_batch.Put("%08d" % self.train_no, serialized)
self.train_no += 1
elif mode == 'valid':
self.valid_batch.Put("%08d" % self.valid_no, serialized)
self.valid_no += 1
elif mode == 'test':
self.test_batch.Put("%08d" % self.test_no, serialized)
self.test_no += 1
def findCorrespondencesMax_createInputs(candidates, normalized_images, cycles,
channels, nbit, n_threads, radius):
"""INPUTS:
candidates=binary max masks
normalized_images=normalized images,
cycles=number of sequencing cycles,
nbit= encoding"""
se = ball(radius)
im_th = np.zeros(normalized_images.shape).astype(np.float64)
for cycle in range(cycles):
for ch in range(2, channels):
im_th[cycle, ch, :, :, :] = white_tophat(
normalized_images[cycle, ch, :, :, :], se)
candidates[candidates == np.iinfo(nbit).max] = 1
res = Parallel(n_jobs=n_threads)(
delayed(Execute_Correspondences_CreateInputs)(
candidates, normalized_images, im_th, cycle, channels, nbit)
for cycle in range(cycles))
max_df = pd.concat([res[i]['max_df'] for i in range(len(res))],
ignore_index=True)
inputs_df = pd.concat([res[i]['inputs_df'] for i in range(len(res))],
ignore_index=True)
inputs_df = inputs_df[[
'cycle', 'ch', 'x', 'y', 'z', 'Intensities_window_5x5'
]]
return {'inputs_df': inputs_df, 'max_df': max_df}
def bw_transform(img_src): # To finish
tval = threshold_otsu(img_src)
bn_src = (img_src > tval)
# Reducing horizontal and vertical lines
r_vertical = rectangle(bn_src.shape[1], 1)
r_horizontal = rectangle(1, bn_src.shape[0])
# Reducing horizontal lines
bn_src = white_tophat(bn_src, r_horizontal)
# Reducing vertical lines
bn_src = white_tophat(bn_src, r_vertical)
# Reducing salt
bn_src = remove_small_objects(bn_src, connectivity=2, min_size=9)
return bn_src
def _segment_plate(image_path):
# Učitaj sliku:
car_image = imread(image_path, as_grey=True)
# Postavi piksele da budu vrijednosti izmedu 0 i 255, umjesto izmedu 0 i 1:
car_image = car_image * 255
# White tophat:
wt_car_image = white_tophat(car_image, disk(10))
# U binary:
block_size = 105
threshold = threshold_local(wt_car_image, block_size, offset=-30)
binary_image = wt_car_image > threshold
# Dilation:
dilated_image = dilation(binary_image, disk(2))
label_image = label(dilated_image)
plate_like_objects = []
for region in regionprops(label_image):
min_row, min_col, max_row, max_col = region.bbox
aspect_ratio = (max_col - min_col) / (max_row - min_row)
if (10000 > region.area > 2000) and (2.2 < aspect_ratio < 4.5):
plate_like_objects.append(car_image[min_row:max_row,
min_col:max_col])
return plate_like_objects
def rolling_disk(image, radius=50, light_bg=False):
from skimage.morphology import white_tophat, black_tophat, disk
str_el = disk(radius)
print(image.shape)
if light_bg:
return black_tophat(image, str_el)
else:
return white_tophat(image, str_el)
def subtract_background(image, radius, light_bg=False):
# you can also use 'ball' here to get a slightly smoother result at the
# cost of increased computing time
str_el = disk(radius)
if light_bg:
return black_tophat(image, str_el)
else:
return white_tophat(image, str_el)
def run(img):
if len(img.shape) > 2 and img.shape[2] == 4:
img = color.rgba2rgb(img)
if len(img.shape) == 2:
img = color.gray2rgb(img)
img = color.rgb2gray(img)
img = white_tophat(img, disk(1))
return to_base64(img)
def morphology_filter(self, image):
morph_img = np.asarray(image)
kernel_dil = np.ones((3, 3), np.uint8)
if len(morph_img.shape) >= 3:
try:
#print("***converting to grayscale***")
morph_for_ski = img_as_float(color.rgb2gray(morph_img))
# binearisation
image_binary = morph_for_ski > 0.001
morph_for_ski[np.logical_not(image_binary)] = 0
# binearization - Black tophat
wth = white_tophat(morph_img)
ch_one = wth[:, :, 0] > 5
ch_two = wth[:, :, 1] > 5
ch_three = wth[:, :, 2] > 5
morph_for_ski[ch_one] = 0
morph_for_ski[ch_two] = 0
morph_for_ski[ch_three] = 0
return (morph_for_ski*255).astype(np.uint8)
# image_binary = morphology.opening(morph_for_ski)
# image_binary = cv2.dilate(np.float32(image_binary), kernel_dil, iterations=3)
# out_skel = skeletonize(image_binary, method='lee')
except ValueError:
print(ValueError)
print("Can't convert this image to grayscale")
return image
# good work with denoising, gamma, sv from hsv and knee filtering
# this is much more precisly but, takes much longer time
else:
try:
print("***Collecting data..***")
morph_for_ski = img_as_float(morph_img)
# binearisation
image_binary = morph_for_ski > 0.001
morph_for_ski[np.logical_not(image_binary)] = 0
return (morph_for_ski * 255).astype(np.uint8)
#image_binary = morphology.opening(morph_for_ski)
#image_binary = cv2.dilate(np.float32(image_binary), kernel_dil, iterations=3)
#out_skel = skeletonize(image_binary, method='lee')
#return out_skel
except ValueError:
print(ValueError)
print("Can't convert this image to grayscale")
return image
def mbi_feature(image_name, output=None, postprocess=True, smooth_factor=None):
MBI_THRESHOLD = 5.5
ds = gdal.Open(image_name)
image = ds.ReadAsArray()
geotran = ds.GetGeoTransform()
ulx = geotran[0]
uly = geotran[3]
cell_width = geotran[1]
cell_height = geotran[5]
out_srs = osr.SpatialReference()
out_srs.ImportFromEPSG(4326)
out_srs_wkt = out_srs.ExportToWkt()
# out_cell_width = block * cell_width
# out_cell_height = block * cell_height
ds = None
image = np.moveaxis(image, 0, -1) # rows, columns, channels
# calculate brightness as a local max
brightness = calc_stats(image, ["max"], 2)
# could be useful to smooth...
if smooth_factor:
brightness = gaussian(brightness, smooth_factor)
# a set of linear structural elements
# for the white tophat transformation
# dirs = [45, 90, 135, 180]
se_sizes = [5, 9, 13, 19, 23, 27]
se_set = get_se_set(se_sizes)
# 'white' top-hat transformation
# in this case, white top-hat is the brightness image minus morphological opening
mean_w_tophats = []
for s in se_set: # for each size in the structural element set
w_tophats = []
for k in s: # for each direction kernel in the structural element set for this size
# directional top hat transformation using linear SE
w_tophats.append(white_tophat(brightness, k))
mean_w_tophat = calc_stats(w_tophats, ['mean'], 0)
mean_w_tophats.append(mean_w_tophat)
th_dmp = []
th_idx = 0
# calculate the differential morphological profile
while th_idx + 1 < len(mean_w_tophats):
th_dmp.append(
np.absolute(mean_w_tophats[th_idx + 1] - mean_w_tophats[th_idx]))
th_idx += 1
mbi = calc_stats(np.array(th_dmp), ['mean'], 0)
if postprocess:
mbi = np.where(mbi >= MBI_THRESHOLD, 1, 0)
if output:
# out_geotran = (out_ulx, out_cell_width, 0, out_uly, 0, out_cell_height)
write_geotiff(output, mbi, geotran, out_srs_wkt)
return np.array(mbi)
def getScalemask(grayimg):
"""
create a mask to capture scale marks
note: this is not pure - text labels are also captured
grayimg = input 2D image array
"""
selem = morphology.disk(1)
return morphology.white_tophat(grayimg, selem)
def white_tophat(img):
orig_img = util.img_as_ubyte(img)
image = orig_img.copy()
image[340:350, 200:210] = 255
image[100:110, 200:210] = 0
selem = morphology.disk(6)
eroded = morphology.white_tophat(image, selem)
plot_comparison(orig_img, eroded, 'White Tophat')
plt.tight_layout()
plt.show()
def BackgroundSubtraction_runnable(cls,
image,
strucsize=50,
disk=disk,
white_tophat=white_tophat):
struc = disk(strucsize)
image = white_tophat(image, struc)
return image
def process_tophat(self, im_denoise):
"""
Process image using tophat, closing, then dilation
"""
im_wtophat = morphology.white_tophat(im_denoise,
selem=morphology.disk(
self.C['WTOPHAT_FILTER_SZ']))
im_closed = morphology.closing(im_wtophat)
im_processed = morphology.dilation(im_closed)
return im_processed
def calc_mbi(raster_input, s_min, s_max, s_delta):
""" Calculates morphological building index (MBI) per
Huang et al. (2015)
Assumes use of linear structuring element
MBI = sum(dxs)(whitetop_hat(morphological profiles))/(D*S)
d = degrees (0,45, 90, 135)
D = 4
s = size of linear structuring element (pixels)
S = (s_max-s_min)/s_delta + 1
Parameters
------------
raster_input : 3-dim raster (tif) stack for calculating brightness
s_min, s_max, s_delta: numerical input (integer)
Returns
------------
Calculated values for MBI in a numpy array
"""
# Calculate brightness for cloud masked stack
brightness = np.nanmax(raster_input, axis=0)
# Cap brightness values at a max of 1. Replace all values greater than 1 with a value of 1
brightness_cap = np.where(brightness > 1, 1, brightness)
# Initialize inputs for MBI calculation
selem = selemline(0, 0)
w_tophat_array_sum = white_tophat(brightness_cap, selem)
# Loop and sum black tophat morphological profiles for MSI calculation
for i in range(s_min, s_max + s_delta, s_delta):
for x in range(0, 4):
selem = selemline(i, 45 * x)
w_tophat = white_tophat(brightness_cap, selem)
w_tophat_array_sum = w_tophat_array_sum.__add__(w_tophat)
D = 4
S = ((s_max - s_min) / s_delta) + 1
mbi = w_tophat_array_sum / (D * S)
return mbi
def filterWithWhiteTophat(image, radius):
"""Wrapper for skimage's white tophat filter
Parameters
----------
image : np.ndarray
Image to be filtered.
radius: int
Radius of the morphological disk.
"""
selem = disk(radius)
return white_tophat(image, selem)
def stara(smap,
circle_radius: u.deg = 100 * u.arcsec,
median_box: u.deg = 10 * u.arcsec,
threshold=6000,
limb_filter: u.percent = None):
"""
A method for automatically detecting sunspots in white-light data using morphological operations.
Parameters
----------
smap : `sunpy.map.GenericMap`
The map to apply the algorithm to.
circle_radius : `astropy.units.Quantity`, optional
The angular size of the structuring element used in the
`skimage.morphology.white_tophat`. This is the maximum radius of
detected features.
median_box : `astropy.units.Quantity`, optional
The size of the structuring element for the median filter, features
smaller than this will be averaged out.
threshold : `int`, optional
The threshold used for detection, this will be subject to detector
degradation. The default is a reasonable value for HMI continuum images.
limb_filter : `astropy.units.Quantity`, optional
If set, ignore features close to the limb within a percentage of the
radius of the disk. A value of 10% generally filters out false
detections around the limb with HMI continuum images.
"""
data = invert(smap.data)
# Filter things that are close to limb to reduce false detections
if limb_filter is not None:
hpc_coords = sunpy.map.all_coordinates_from_map(smap)
r = np.sqrt(hpc_coords.Tx**2 + hpc_coords.Ty**2) / (
smap.rsun_obs - smap.rsun_obs * limb_filter)
data[r > 1] = np.nan
# Median filter to remove detections based on hot pixels
m_pix = int((median_box / smap.scale[0]).to_value(u.pix))
med = median(data, square(m_pix), behavior="ndimage")
# Construct the pixel structuring element
c_pix = int((circle_radius / smap.scale[0]).to_value(u.pix))
circle = disk(c_pix / 2)
finite = white_tophat(med, circle)
finite[np.isnan(finite)] = 0 # Filter out nans
return finite > threshold
def white_tophat():
image = io.imread("pics/Cosmos.jpg")
white = morphology.white_tophat(image)
fig, (ax0, ax1) = plt.subplots(nrows=1, ncols=2, figsize=(8, 3),
sharex=True, sharey=True)
ax0.imshow(image, cmap=plt.cm.gray)
ax0.axis('off')
ax1.imshow(white, cmap=plt.cm.gray)
ax1.axis('off')
plt.show()
def extract_leaf_stem(image, maxdisksize=30, minstempixels=100):
width,height,depth = np.shape(image)
X = image.reshape(width*height,depth)
clast = KMeans(n_clusters=2, n_jobs=-2)
clast.fit(X)
lbls = np.array(clast.labels_).reshape(width,height)
if np.mean(lbls[width/2-10:width/2+10,height/2-10:height/2+10])>0.5:
rightlbl=1
else:
rightlbl=0
lbls = (lbls==rightlbl).astype(np.int16)
im_flt=white_tophat(lbls,disk(maxdisksize))
# cnts=measure.find_contours(im_flt, 0.8)
flb=measure.label(im_flt,connectivity=2)
props=measure.regionprops(flb)
cc = [item['inertia_tensor_eigvals'][0]/item['inertia_tensor_eigvals'][1] if item['inertia_tensor_eigvals'][1]>0.0 and item['area']>=minstempixels else 0.0 for item in props]
res = lbls.copy()
if any(cc):
indmax = np.nanargmax(cc)
stem = (flb==indmax+1)
res[stem]=0
return res
def vesselEnhance(self, array=numpy.empty(0)):
"""
@method vesselEnhance
@param array {numpy array} the array the operation is carried out
on, default is the image_array.
"""
self.__printStatus("Vessel enhancement...");
if not array.any():
array = self.image_array
# disk shaped mask with radius 8
disk_shape = disk(8)
# the complimentary image is saved to hc:
array = numpy.uint8(array)
hc = 255 - array
# Top Hat transform
# https://en.wikipedia.org/wiki/Top-hat_transform
# White top hat is defined as the difference between
# the opened image and the original image.
# in this case the starting image is the complimentary image `hc`
self.image_array = white_tophat(hc, selem=disk_shape) * self.mask
self.__printStatus("[done]", True)
return self
# mostly removed. # # White tophat # ============ # # The ``white_tophat`` of an image is defined as the *image minus its # morphological opening*. This operation returns the bright spots of the # image that are smaller than the structuring element. # # To make things interesting, we'll add bright and dark spots to the image: phantom = orig_phantom.copy() phantom[340:350, 200:210] = 255 phantom[100:110, 200:210] = 0 w_tophat = white_tophat(phantom, selem) plot_comparison(phantom, w_tophat, 'white tophat') ###################################################################### # As you can see, the 10-pixel wide white square is highlighted since it is # smaller than the structuring element. Also, the thin, white edges around # most of the ellipse are retained because they're smaller than the # structuring element, but the thicker region at the top disappears. # # Black tophat # ============ # # The ``black_tophat`` of an image is defined as its morphological **closing # minus the original image**. This operation returns the *dark spots of the # image that are smaller than the structuring element*.
rots = np.random.uniform(0,360,10).astype(int) #10 random rotations
for i, rot in enumerate(rots):
im_size = img_org.shape[0]
if (np.random.rand() > 0.5):
if (np.random.rand() > 0.5):
img_org = cv2.flip(img_org,0)
else:
img_org = cv2.flip(img_org,1)
scale = np.random.uniform(0.7,1.3)
mat = cv2.getRotationMatrix2D((im_size / 2, im_size / 2), rot, scale=scale)
img_rotated = cv2.warpAffine(img_org, mat, (im_size, im_size), flags=cv2.INTER_LINEAR, borderValue=(255,255,255))
img_out = np.zeros((img_rotated.shape[0], img_rotated.shape[1], 3), dtype=np.uint8)
img_orig = img_rotated[:,:,0]
img_btop = 255-black_tophat(img_orig, selem)
img_wtop = 255-white_tophat(img_orig, selem)
img_out[:, :, 1] = img_btop
img_out[:, :, 2] = img_wtop
img_rotated = img_out
if show:
cv2.imshow('Rot_' + str(i), img_rotated)
writeImg(outPath, img_rotated, kind, file, i+1) #Original
if show:
cv2.waitKey(20000 )
lineNum += 1
if (lineNum > 20):
break
# get 'structureness' for each channel at each pixel
R = (K1 / K2)**2
S = K1**2 + K2**2
# conservative threshold
R_thresh = ma.median(R)
targets = (R < R_thresh).filled(0)
# erode based on scale space
erode_selem = disk(sigma+1)
dilate_selem = disk(max(sigma-1, 1))
min_size = dilate_selem.sum()*2
targets = binary_erosion(targets, selem=erode_selem)
#targets = binary_dilation(targets, selem=dilate_selem)
targets = white_tophat(targets, selem=disk(sigma))
labeled, n_labels = label(targets, background=0,
connectivity=1, return_num=True)
sizes = np.bincount(labeled.flatten())
rankings = np.argwhere(sizes >= min_size)
# labels for the N largest regions
N = rankings.size - 1 # number of regions to view
largest = rankings[1:] # everything except background
candidates = np.zeros_like(labeled)
for region_label in largest:
candidates = np.logical_or(candidates, labeled == region_label)
def convert_test_data(self, data_set_folder, min_pixel, test_db_name, test_output_pickle_path,
inverse, channel_no = 1):
self.remove_folder(test_db_name)
test_db = leveldb.LevelDB(test_db_name)
pickleTestX = test_output_pickle_path + "/testX_size_" + str(min_pixel) + ".pickle"
pickleFileNames = test_output_pickle_path + "/fileNames.pickle"
if not os.path.exists(test_output_pickle_path):
os.makedirs(test_output_pickle_path)
numberofImages = 0
datum = caffe.proto.caffe_pb2.Datum()
datum.channels = channel_no
datum.width = min_pixel
datum.height = min_pixel
test_batch = leveldb.WriteBatch()
print "Load test dataset from image files"
for fileNameDir in os.walk(data_set_folder):
for index, fileName in enumerate(fileNameDir[2]):
if fileName[-5:] != ".JPEG":
continue
numberofImages += 1
imageSize = min_pixel * min_pixel
num_rows = numberofImages # one row for each image in the test dataset
batch_size = 10000
data_size = min(batch_size, numberofImages)
testX = numpy.zeros((data_size, channel_no, imageSize), dtype=numpy.uint8)
files = []
db_index = 0
pickle_index = 0
batch_no = 1
print "Reading images"
for fileNameDir in os.walk(data_set_folder):
for index, fileName in enumerate(fileNameDir[2]):
if fileName[-5:] != ".JPEG":
continue
nameFileImage = "{0}{1}{2}".format(fileNameDir[0], os.sep, fileName)
org_image = Image.open(nameFileImage)
files.append(fileName)
image = org_image.resize((min_pixel, min_pixel), Image.ANTIALIAS)
"""
print fileName
plt.figure(1, figsize=(1, 1), dpi=100)
plt.gray();
plt.subplot(1, 1, 1)
plt.imshow(image)
plt.show()
"""
if inverse:
image_ubyte = 255 - img_as_ubyte(image)
else:
image_ubyte = img_as_ubyte(image)
if channel_no > 1:
selem = disk(6)
w_tophat = white_tophat(image_ubyte, selem)
b_tophat = black_tophat(image_ubyte, selem)
datum.data = image_ubyte.tostring() + w_tophat.tostring() + b_tophat.tostring()
image_output = numpy.concatenate((image_ubyte, w_tophat, b_tophat), axis=1)
else:
datum.data = image_ubyte.tostring()
image_output = image_ubyte
test_batch.Put("%08d" % db_index, datum.SerializeToString())
testX[pickle_index] = numpy.reshape(image_output, (channel_no, imageSize))
db_index += 1
pickle_index += 1
if db_index % 1000 == 0:
test_db.Write(test_batch, sync = True)
del test_batch
test_batch = leveldb.WriteBatch()
print 'Processed %i test images.' % db_index
if pickle_index % batch_size == 0:
pickle_file_name = pickleTestX + "_" + str(batch_no)
with open(pickle_file_name,'wb') as fp:
cPickle.dump(testX, fp)
print "pickled %s" % pickle_file_name
data_size = min(batch_size, numberofImages - batch_size * batch_no)
testX = numpy.zeros((data_size, channel_no, imageSize), dtype=numpy.uint8)
batch_no += 1
pickle_index = 0
report = [int((j+1)*num_rows/20.) for j in range(20)]
if db_index in report: print numpy.ceil(db_index *100.0 / num_rows), "% done"
# Write last batch of images
if db_index % 1000 != 0:
test_db.Write(test_batch, sync = True)
if pickle_index % batch_size > 0:
pickle_file_name = pickleTestX + "_" + str(batch_no)
with open(pickle_file_name,'wb') as fp:
cPickle.dump(testX, fp)
print "pickled %s" % pickle_file_name
with open(pickleFileNames,'wb') as fp:
cPickle.dump(files, fp)
print 'Processed a total of %i images.' % db_index
import numpy as np
from PIL import Image
from skimage.morphology import erosion, dilation, opening, closing
from skimage.morphology import black_tophat, white_tophat
from skimage.morphology import disk, diamond, rectangle, square, star
def my_imshow(arr, filter_name):
f,ax = plt.subplots()
ax.set_title(filter_name)
#ax.axis('off')
ax.set_adjustable('box-forced')
ax.imshow(arr, cmap=cm.gray, interpolation='none')
plt.show()
#Import an image.
image = np.array(Image.open('Butterfly.jpg'))
my_imshow(image, 'original')
#White tophat
fly = img.copy()
fly[340:350, 200:210] = 255
fly[100:110, 200:210] = 0
w_tophat = white_tophat(fly, selem)
my_imshow(w_tophat, 'white-tophat')
#Black tophat
b_tophat = black_tophat(fly, selem)
my_imshow(b_tophat, 'black-tophat')
#t.reset();
#res2 = open(img, structureElement('Disk', (30,30)).astype('bool'));
#t.printElapsedTime('mahotas');
t.reset();
se = structureElement('Disk', (15,15)).astype('uint8');
res2 = cv2.morphologyEx(img, cv2.MORPH_OPEN, se)
t.printElapsedTime('opencv');
#from skimage.morphology import disk
#se = disk(10);
t.reset();
#res3 = opening(img.astype('uint16'), se);
res3 = white_tophat(img.astype('uint16'), se);
t.printElapsedTime('skimage');
t.reset();
x = (res3);
print (x.min(), x.max())
plotTiling(numpy.dstack((10 * img, 10 * (img - res2))))
#seems not to work correctly -> scipy gets very slow for larger images (??)
"""