def load_file(self, data_dir, file_name, num_channels=3):
try:
self.data_dir = data_dir
self.file_name = file_name
self.image_arr = imgutil.get_image_as_array(os.path.join(self.data_dir, self.file_name),
channels=num_channels)
except Exception:
try:
self.image_arr = imgutil.get_image_as_array(os.path.join(self.data_dir, self.file_name),
channels=3 if num_channels == 1 else 1)
except Exception as e1:
print('Error Loading file: ' + self.file_name)
print(str(e1))
def load_mask(self, mask_dir=None, fget_mask=None, erode=False, channels=1):
try:
mask_file = fget_mask(self.file_name)
self.mask = imgutil.get_image_as_array(os.path.join(mask_dir, mask_file), channels)
if erode:
self.mask = cv2.erode(self.mask, kernel=fu.get_chosen_mask_erode_kernel(), iterations=5)
except Exception as e:
print('Fail to load mask: ' + str(e))
def _get_image_obj(self, img_file=None):
img_obj = Image()
img_obj.load_file(data_dir=self.image_dir, file_name=img_file)
if self.mask_getter is not None:
img_obj.load_mask(mask_dir=self.mask_dir,
fget_mask=self.mask_getter)
if self.truth_getter is not None:
img_obj.load_ground_truth(gt_dir=self.truth_dir,
fget_ground_truth=self.truth_getter)
img_obj.working_arr = img_obj.image_arr[:, :, 1]
img_obj.apply_clahe()
img_obj.apply_mask()
sup, res = 20, 235
img_obj.extra['unet'] = iu.get_image_as_array(
self.unet_dir + sep + img_obj.file_name.split('.')[0] +
self.input_image_ext, 1)
img_obj.extra['indices'] = list(
zip(*np.where((img_obj.extra['unet'] >= sup)
& (img_obj.extra['unet'] <= res))))
img_obj.extra['fill_in'] = np.zeros_like(img_obj.working_arr)
img_obj.extra['fill_in'][img_obj.extra['unet'] > res] = 1
img_obj.extra['mid_pix'] = img_obj.extra['unet'].copy()
img_obj.extra['mid_pix'][img_obj.extra['mid_pix'] < sup] = 0
img_obj.extra['mid_pix'][img_obj.extra['mid_pix'] > res] = 0
img_obj.extra['gt_mid'] = img_obj.ground_truth.copy()
img_obj.extra['gt_mid'][img_obj.extra['unet'] > res] = 0
img_obj.extra['gt_mid'][img_obj.extra['unet'] < sup] = 0
# <PREP1> Segment with a low threshold and get a raw segmented image
raw_estimate = img_obj.extra['unet'].copy()
raw_estimate[raw_estimate > sup] = 255
raw_estimate[raw_estimate <= sup] = 0
# <PREP2> Clear up small components(components less that 10px)
raw_estimate = iu.remove_connected_comp(raw_estimate.squeeze(), 10)
# <PREP3> Skeletonize binary image
seed = raw_estimate.copy()
seed[seed == 255] = 1
seed = skeletonize(seed).astype(np.uint8)
# <PREP4> Come up with a grid mask to select few possible pixels to reconstruct the vessels from
sk_mask = np.zeros_like(seed)
sk_mask[::int(0.6 * self.patch_shape[0])] = 1
sk_mask[:, ::int(0.6 * self.patch_shape[0])] = 1
# <PREP5> Apply mask and save seed
img_obj.extra['seed'] = seed * sk_mask * 255
return img_obj
def run_all(self,
Dirs=None,
fget_mask=None,
fget_gt=None,
params_combination=[],
save_images=False):
if os.path.isdir(self.out_dir) is False:
os.makedirs(self.out_dir)
self.writer = open(self.out_dir + os.sep + "segmentation_result.csv",
'w')
self.writer.write('ITR,FILE_NAME,F1,PRECISION,RECALL,ACCURACY,'
'SK_THRESHOLD,'
'ALPHA,'
'ORIG_CONTRIB,'
'SEG_THRESHOLD\n')
for file_name in os.listdir(Dirs['images']):
print('File: ' + file_name)
img_obj = SegmentedImage()
img_obj.load_file(data_dir=Dirs['images'], file_name=file_name)
img_obj.working_arr = img_obj.image_arr[:, :, 1]
img_obj.apply_clahe()
img_obj.res['orig'] = img_obj.working_arr
img_obj.working_arr = imgutils.get_image_as_array(
Dirs['segmented'] + sep + file_name + '.png', channels=1)
img_obj.load_mask(mask_dir=Dirs['mask'],
fget_mask=fget_mask,
erode=True)
img_obj.load_ground_truth(gt_dir=Dirs['truth'],
fget_ground_truth=fget_gt)
img_obj.apply_mask()
img_obj.generate_lattice_graph()
arr = img_obj.working_arr.copy()
for i in range(arr.shape[0]):
for j in range(arr.shape[1]):
if img_obj.mask[i, j] == 0:
arr[i, j] = 255
img_obj.working_arr = arr
for params in params_combination:
img_obj.generate_skeleton(
threshold=int(params['sk_threshold']))
self._run(img_obj=img_obj,
params=params,
save_images=save_images)
self.writer.close()
def load_ground_truth(self, gt_dir=None, fget_ground_truth=None, channels=1):
try:
gt_file = fget_ground_truth(self.file_name)
self.ground_truth = imgutil.get_image_as_array(os.path.join(gt_dir, gt_file), channels)
except Exception as e:
print('Fail to load ground truth: ' + str(e))