def load_images(self, image_dir):
image_files = general_utils.get_all_files(image_dir)
for image_file in image_files:
parts = image_file[:-4].split("_")
if parts[1] == "draw":
continue
idx = int(parts[0]) - 1
self.blocks[idx].reload_image(source=image_dir + "/" + image_file)
def get_test_image():
chromosome_dir = data_dir + "/chromosome"
chromosome_ids = [
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20,
21, 22, "x", "y"
]
idx = np.random.randint(len(chromosome_ids))
chromosome_dir = chromosome_dir + "/" + str(chromosome_ids[idx])
image_files = general_utils.get_all_files(chromosome_dir)
image_file = image_files[np.random.randint(len(image_files))]
test_image = cv2.imread(chromosome_dir + "/" + image_file, 0)
return test_image
def test_generate_chromosome_cluster(self):
image_dir = data_dir + "/chromosome/1"
image_files = general_utils.get_all_files(image_dir)
num_chromosome = 15
contours = list()
chromosomes = list()
chosen_files = list()
# error_case = ['/home/lntk/Desktop/Karyotype/data/chromosome/1/xx_karyotype_225_0.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_213_1.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_190_1.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xx_karyotype_064_0.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xx_karyotype_017_0.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_055_0.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_208_0.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_266_1.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_132_0.bmp',
# '/home/lntk/Desktop/Karyotype/data/chromosome/1/xy_karyotype_233_1.bmp']
for _ in range(num_chromosome):
""" Randomly get chromosome image from directory """
idx = np.random.randint(len(image_files))
chromosome = image_utils.read_image(image_dir + "/" +
image_files[idx])
chosen_files.append(image_dir + "/" + image_files[idx])
# for image_file in error_case:
# chromosome = image_utils.read_image(image_file)
""" Resize """
chromosome = cv2.resize(chromosome, (64, 64))
""" Rotate """
angle = np.random.randint(90)
rotated_chromosome = image_utils.rotate_image(chromosome, angle)
""" Extract contour """
contour = image_utils.get_chromosome_contour(rotated_chromosome)
contours.append(contour)
chromosomes.append(rotated_chromosome)
print(chosen_files)
contours, boxes, initial_boxes = chromosome_utils.generate_chromosome_cluster(
contours)
for contour in contours:
print(contour.shape)
# silhouette_image = chromosome_utils.get_chromosome_silhouette(contours, boxes)
cluster_image = chromosome_utils.get_chromosome_cluster_image(
boxes, initial_boxes, chromosomes)
image_utils.show_multiple_images([cluster_image])
def directory_to_images_files_and_labels(directory, verbose=False):
if verbose:
print("Processing " + directory)
relative_label_dirs = [x for x in next(os.walk(directory))[1]
] # get label names, such as "1", "2", ....
label_to_id = {
key: value
for (value, key) in enumerate(relative_label_dirs)
} # mapping label to id (0, 1, ...)
id_to_label = {
key: value
for (key, value) in enumerate(relative_label_dirs)
}
if verbose:
print(label_to_id)
image_ids, label_ids = list(), list()
for label in relative_label_dirs:
label_id = int(label_to_id[label])
label_dir = directory + "/" + label
if verbose:
print("Processing " + label_dir)
image_files = general_utils.get_all_files(label_dir)
for image_file in image_files:
image_file = label_dir + "/" + image_file
image_ids.append(image_file)
label_ids.append(label_id)
if verbose:
print("#Images so far: ", len(image_ids))
if verbose:
print("10 first image ids:", image_ids[:10])
print("10 first label ids:", label_ids[:10])
print("10 first label:", [id_to_label[idx] for idx in label_ids[:10]])
return image_ids, label_ids, id_to_label, label_to_id
def load_images(self, image_dir):
image_files = general_utils.get_all_files(image_dir)
for image_file in image_files:
idx = int(image_file[:-4]) - 1
self.blocks[idx].reload_image(source=image_dir + "/" + image_file)
def process_karyotyping_images(directory,
chromosome_type="xx",
debug=False,
load_karyotype_info=False,
max_p=None,
need_confirm=False):
last_chromosome = chromosome_type[1]
chromosome_ids = [
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20,
21, 22, "x", "y"
]
save_directory = directory + "/" + chromosome_type + "/chromosome"
karyotype_directory = directory + "/" + chromosome_type
if debug:
print("Save directory: " + save_directory)
print("Karyotype directory: " + karyotype_directory)
general_utils.create_directory(save_directory)
if not load_karyotype_info:
# Create directories to store output images
for idx in chromosome_ids:
general_utils.create_directory(save_directory + "/" + str(idx))
karyotypes = dict()
image_files = general_utils.get_all_files(karyotype_directory)
for image_file in image_files:
if not image_file.endswith(".bmp"):
continue
if debug:
print(image_file)
image = cv2.imread(karyotype_directory + "/" + image_file, 0)
# ret, thresh = threshold.partial_otsu_threshold(image, minval=0, maxval=255, dark_background=False)
ret, thresh = cv2.threshold(image, 254, 255, cv2.THRESH_BINARY_INV)
# if debug:
# image_utils.show_image(thresh)
_, contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
num_contour = len(contours)
# Find area threshold - taking the 46-th largest area
areas = list()
for idx in range(num_contour):
is_outer_contour = hierarchy[0][idx][3] == -1
if is_outer_contour:
area = cv2.contourArea(contours[idx])
areas.append(area)
areas.sort(reverse=True)
if len(areas) < 46:
print("Wrong at " + image_file)
continue
area_threshold = areas[45] - 0.01
chosen_contours = list()
for idx in range(num_contour):
contour = contours[idx]
if cv2.contourArea(contour) < area_threshold:
continue
chosen_contours.append(contour)
if len(chosen_contours) < 46:
print("Wrong at " + image_file)
continue
contour_info = list()
for contour in chosen_contours:
x, y, w, h = cv2.boundingRect(contour)
contour_info.append([x, y, w, h, contour])
def sorted_by(a, b):
x_a, y_a, w_a, h_a, _ = a
x_b, y_b, w_b, h_b, _ = b
if (y_a + h_a) < y_b:
return -1
if x_a < x_b:
return -1
return 1
cmp = functools.cmp_to_key(sorted_by)
contour_info.sort(key=cmp)
if need_confirm:
rgb_image = np.stack((image, ) * 3, axis=-1)
image_utils.show_image(image_utils.get_image_with_contours(
rgb_image, chosen_contours, thickness=-1),
cmap=None)
user_input = input()
if "yes".startswith(user_input):
karyotypes[image_file] = contour_info
else:
print("Skipping: " + image_file)
else:
rgb_image = np.stack((image, ) * 3, axis=-1)
image_utils.show_image(image_utils.get_image_with_contours(
rgb_image, chosen_contours, thickness=-1),
cmap=None)
karyotypes[image_file] = contour_info
# Save data in case of bugs
pickle.dump(
karyotypes,
open(directory + "/" + chromosome_type + "_karyotype_info.data",
'wb'))
else:
with open(directory + "/" + chromosome_type + "_karyotype_info.data",
'rb') as f:
karyotypes = pickle.load(f)
# Find maximum perimeter
if max_p is None:
max_p = -1
for idx in karyotypes.keys():
contour_info = karyotypes[idx]
if len(contour_info) != 46:
print("Skipping:" + idx)
continue
con_1 = contour_info[0][4]
con_2 = contour_info[1][4]
p_1 = cv2.arcLength(con_1, True)
p_2 = cv2.arcLength(con_2, True)
max_p = max(p_1, max_p)
max_p = max(p_2, max_p)
if debug:
print("Max p: " + str(max_p))
# Resize all images according to max perimeter
for image_file in karyotypes.keys():
image = cv2.imread(karyotype_directory + "/" + image_file, 0)
contour_info = karyotypes[image_file]
if debug:
print(image_file)
if len(contour_info) != 46:
print("Skipping:" + image_file)
continue
# Get max perimeter of two chromosome 1
con_1 = contour_info[0][4]
con_2 = contour_info[1][4]
p_1 = cv2.arcLength(con_1, True)
p_2 = cv2.arcLength(con_2, True)
local_max_p = max(p_1, p_2)
# Get scale according to above local max
scale = max_p * 1.0 / local_max_p
print(scale)
component_size = 512
counter = 1
pixel_open = 0
for idx in range(len(contour_info)):
x, y, w, h, contour = contour_info[idx]
# open the bounding box a little bit
x -= pixel_open
y -= pixel_open
w += pixel_open
h += pixel_open
# create a white image of size 512
white_image = 255 - np.zeros(shape=(component_size,
component_size))
# calculate corresponding top-left point in white image
new_x = int((component_size - w) / 2)
new_y = int((component_size - h) / 2)
# copy bounding box patch from karyotype image into white image
white_image[new_y:(new_y + h),
new_x:(new_x + w)] = image[y:(y + h), x:(x + w)]
# rescale chromosome image
white_image = cv2.resize(
white_image,
(int(component_size * scale), int(component_size * scale)))
white_image = white_image.astype('uint8')
# get the center patch of size 512
white_image = image_utils.get_center_sub_image(white_image,
size=component_size)
# get current chromosome name (1, 2, ..., x, y)
chromosome_name = str(chromosome_ids[int(idx / 2)])
# the last chromosome name depends on what karyotype image (xx or xy)
if idx == 45:
chromosome_name = last_chromosome
# save the patch contain only 1 chromosome
cv2.imwrite(
save_directory + "/" + chromosome_name + "/" +
chromosome_type + "_" + image_file + "_" + str(idx),
white_image)
counter += 1