def __data_generation(self, batch_names):
# Initialization of the batches
X = np.empty((self.batch_size, self.size[0], self.size[1], 3))
# Change for using only 1 class "FIBERS" as output , last dimension to 1
y = np.empty((self.batch_size, self.size[0], self.size[1], 1))
for i, name in enumerate(batch_names):
X[i, :, :, :] = load_image(self.path,
name) # Load the data for the image
# Load the data for the classification images
if isfile(
join(self.path,
name.replace(self.data_type,
'_FIB' + self.data_type))):
# Load the image of the annotations
label = cv2.imread(
join(self.path,
name.replace(self.data_type,
'_FIB' + self.data_type)), 0)
# Threshold the image (while reading some annotations are not with values [0,255] only)
_, label = cv2.threshold(label, 127, 255, 0)
else:
# Create a label matrix with all zeros if the image was not annotated (doesn't have fibers)
label = np.zeros(self.size[0], self.size[1])
y[i, :, :, 0] = 1 * (
label == 255
) # Classify as 1 if the value is 255 for the "fiber" mask.
# Change for using only 1 class "FIBERS" as output
# y[i,:,:,1] = 1*(label==0) # Classify as 1 if the value is 0 for the "not a fiber" mask.
return X, y
def remove_trash(self):
"""
This function produces a new file PRED2 that removes contours which are too small.
:return:
"""
for num_name, name in enumerate(self.names):
img = load_image(self.path, name)
contours_per_image = self.images_contours[num_name]
mask = np.zeros(img.shape, dtype="uint8")
for num_contour, contour in enumerate(contours_per_image):
if self.classification[num_name][num_contour] == 0:
img = cv2.drawContours(img, contour, -1, 255, -1)
cv2.imwrite(join(self.path, name), img)
def get_contours(self, names):
"""
A function that obtains the contours for each of the prediction images
:return: A list with the contours for each of the images
"""
images_contours = []
label_contours = []
for num_name, name in enumerate(names):
img = load_image(self.path, name)
img = img.astype(np.uint8)
__, thresh = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY_INV)
__, contours_per_image, __ = cv2.findContours(
thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
images_contours.append(contours_per_image)
return images_contours
def save_numbered_fibers(self, save_path):
"""
This function saves the prediction images. For each of the contours a number with its area will appear beside.
:param save_path: Where should these new images should be included
:return:
"""
# For each image get its contours and write the area in a new image
for num_name, name in enumerate(self.names):
img = load_image(self.path, name)
contours_per_image = self.images_contours[num_name]
for num_contour, contour in enumerate(contours_per_image):
cv2.putText(img,
str(self.images_areas[num_name][num_contour]),
(contour[0][0][0], contour[0][0][1]),
fontFace=cv2.FONT_HERSHEY_TRIPLEX,
fontScale=0.8,
color=4,
thickness=1)
cv2.imwrite(join(save_path, name.replace('PRED', 'AREA')), img)
def get_image_size(self):
img = load_image(self.path, self.names[-1])
return img.shape[0]