def get_images_diaretdb(image_dir, preprocess=False):
if preprocess:
limit = 2
grid_size = 8
if not os.path.exists(os.path.join(image_dir, 'ddb1_fundusimages_CLAHE')):
os.mkdir(os.path.join(image_dir, 'ddb1_fundusimages_CLAHE'))
images_ori = glob.glob(os.path.join(image_dir, 'ddb1_fundusimages/*.png'))
images_ori.sort()
# mean brightness.
meanbright = 0.
for img_path in images_ori:
img_name = os.path.split(img_path)[-1].split('.')[0]
mask_path = os.path.join(image_dir, 'ddb1_fundusmask', 'fmask1.tif')
gray = cv2.imread(img_path, 0)
mask_img = cv2.imread(mask_path, 0)
brightness = gray.sum() / (mask_img.shape[0] * mask_img.shape[1] - mask_img.sum() / 255.)
meanbright += brightness
meanbright /= len(images_ori)
# preprocess for images.
for img_path in images_ori:
img_name = os.path.split(img_path)[-1]
mask_path = os.path.join(image_dir, 'ddb1_fundusmask', 'fmask1.tif')
clahe_img = clahe_gridsize(img_path, mask_path, denoise=True, verbose=False, brightnessbalance=meanbright, cliplimit=limit, gridsize=grid_size)
cv2.imwrite(os.path.join(image_dir, 'ddb1_fundusimages_CLAHE', os.path.split(img_path)[-1]), clahe_img)
image_paths = glob.glob(os.path.join(image_dir, 'ddb1_fundusimages_CLAHE/*.png'))
else:
image_paths = glob.glob(os.path.join(image_dir, 'ddb1_fundusimages/*.png'))
image_paths.sort()
return image_paths
def get_images(image_dir, preprocess='0', phase='train'):
if phase == 'train' or phase == 'eval':
setname = 'TrainingSet'
elif phase == 'test':
setname = 'TestingSet'
limit = 2
grid_size = 8
if not os.path.exists(os.path.join(image_dir, 'Images_CLAHE' + preprocess)):
os.mkdir(os.path.join(image_dir, 'Images_CLAHE' + preprocess))
if not os.path.exists(os.path.join(image_dir, 'Images_CLAHE' + preprocess, setname)):
os.mkdir(os.path.join(image_dir, 'Images_CLAHE' + preprocess, setname))
# compute mean brightess
meanbright = 0.
images_number = 0
for tempsetname in ['TrainingSet', 'TestingSet']:
imgs_ori = glob.glob(os.path.join(image_dir, 'OriginalImages/'+ tempsetname + '/*.jpg'))
imgs_ori.sort()
images_number += len(imgs_ori)
# mean brightness.
for img_path in imgs_ori:
img_name = os.path.split(img_path)[-1].split('.')[0]
mask_path = os.path.join(image_dir, 'Groundtruths', tempsetname, 'Mask', img_name + '_MASK.tif')
gray = cv2.imread(img_path, 0)
mask_img = cv2.imread(mask_path, 0)
brightness = gray.sum() / (mask_img.shape[0] * mask_img.shape[1] - mask_img.sum() / 255.)
meanbright += brightness
meanbright /= images_number
imgs_ori = glob.glob(os.path.join(image_dir, 'OriginalImages/' + setname + '/*.jpg'))
preprocess_dict = {'0': [False, False, None], '1': [False, False, meanbright], '2': [False, True, None], '3': [False, True, meanbright], '4': [True, False, None], '5': [True, False, meanbright], '6': [True, True, None], '7': [True, True, meanbright]}
for img_path in imgs_ori:
img_name = os.path.split(img_path)[-1].split('.')[0]
mask_path = os.path.join(image_dir, 'Groundtruths', setname, 'Mask', img_name + '_MASK.tif')
clahe_img = clahe_gridsize(img_path, mask_path, denoise=preprocess_dict[preprocess][0], contrastenhancement=preprocess_dict[preprocess][1], brightnessbalance=preprocess_dict[preprocess][2], cliplimit=limit, gridsize=grid_size)
cv2.imwrite(os.path.join(image_dir, 'Images_CLAHE' + preprocess, setname, os.path.split(img_path)[-1]), clahe_img)
imgs = glob.glob(os.path.join(image_dir, 'Images_CLAHE' + preprocess, setname, '*.jpg'))
imgs.sort()
mask_paths = []
train_number = int(len(imgs) * train_ratio)
eval_number = int(len(imgs) * eval_ratio)
if phase == 'train':
image_paths = imgs[:train_number]
elif phase == 'eval':
image_paths = imgs[train_number:]
else:
image_paths = imgs
mask_path = os.path.join(image_dir, 'Groundtruths', setname)
lesions = ['HardExudates', 'Haemorrhages', 'Microaneurysms', 'SoftExudates', 'Mask']
lesion_abbvs = ['EX', 'HE', 'MA', 'SE', 'MASK']
for image_path in image_paths:
paths = []
name = os.path.split(image_path)[1].split('.')[0]
for lesion, lesion_abbv in zip(lesions, lesion_abbvs):
candidate_path = os.path.join(mask_path, lesion, name + '_' + lesion_abbv + '.tif')
if os.path.exists(candidate_path):
paths.append(candidate_path)
else:
paths.append(None)
mask_paths.append(paths)
return image_paths, mask_paths
def get_images(image_dir,
preprocess=False,
phase='train',
healthy_included=True):
if preprocess:
limit = 2
grid_size = 8
if not os.path.exists(
os.path.join(image_dir, 'ApparentRetinopathy_CLAHE')):
os.mkdir(os.path.join(image_dir, 'ApparentRetinopathy_CLAHE'))
os.mkdir(os.path.join(image_dir, 'NoApparentRetinopathy_CLAHE'))
apparent_ori = glob.glob(
os.path.join(image_dir, 'ApparentRetinopathy/*.jpg'))
noapparent_ori = glob.glob(
os.path.join(image_dir, 'NoApparentRetinopathy/*.jpg'))
apparent_ori.sort()
noapparent_ori.sort()
# mean brightness.
meanbright = 0.
for img_path in apparent_ori + noapparent_ori:
img_name = os.path.split(img_path)[-1].split('.')[0]
mask_path = os.path.join(image_dir, 'GroundTruth', 'MASK',
img_name + '_MASK.tif')
gray = cv2.imread(img_path, 0)
mask_img = cv2.imread(mask_path, 0)
brightness = gray.sum() / (
mask_img.shape[0] * mask_img.shape[1] -
mask_img.sum() / 255.)
meanbright += brightness
meanbright /= len(apparent_ori + noapparent_ori)
# preprocess for apparent.
for img_path in apparent_ori:
img_name = os.path.split(img_path)[-1].split('.')[0]
mask_path = os.path.join(image_dir, 'GroundTruth', 'MASK',
img_name + '_MASK.tif')
clahe_img = clahe_gridsize(img_path,
mask_path,
denoise=True,
verbose=False,
brightnessbalance=meanbright,
cliplimit=limit,
gridsize=grid_size)
cv2.imwrite(
os.path.join(image_dir, 'ApparentRetinopathy_CLAHE',
os.path.split(img_path)[-1]), clahe_img)
# preprocess for noapparent.
for img_path in noapparent_ori:
img_name = os.path.split(img_path)[-1].split('.')[0]
mask_path = os.path.join(image_dir, 'GroundTruth', 'MASK',
img_name + '_MASK.tif')
clahe_img = clahe_gridsize(img_path,
mask_path,
denoise=True,
verbose=False,
brightnessbalance=meanbright,
cliplimit=limit,
gridsize=grid_size)
cv2.imwrite(
os.path.join(image_dir, 'NoApparentRetinopathy_CLAHE',
os.path.split(img_path)[-1]), clahe_img)
apparent = glob.glob(
os.path.join(image_dir, 'ApparentRetinopathy_CLAHE/*.jpg'))
noapparent = glob.glob(
os.path.join(image_dir, 'NoApparentRetinopathy_CLAHE/*.jpg'))
else:
apparent = glob.glob(
os.path.join(image_dir, 'ApparentRetinopathy/*.jpg'))
noapparent = glob.glob(
os.path.join(image_dir, 'NoApparentRetinopathy/*.jpg'))
apparent.sort()
noapparent.sort()
image_paths = []
mask_paths = []
if healthy_included:
imgset = [apparent, noapparent]
else:
imgset = [apparent]
for each in imgset:
train_number = int(len(each) * train_ratio)
eval_number = int(len(each) * eval_ratio)
if phase == 'train':
image_paths.extend(each[:train_number])
elif phase == 'eval':
image_paths.extend(each[train_number:train_number + eval_number])
else:
image_paths.extend(each[train_number + eval_number:])
mask_path = os.path.join(image_dir, 'GroundTruth')
lesions = ['EX', 'HE', 'MA', 'SE', 'MASK']
for image_path in image_paths:
paths = []
name = os.path.split(image_path)[1].split('.')[0]
for lesion in lesions:
candidate_path = os.path.join(mask_path, lesion,
name + '_' + lesion + '.tif')
if os.path.exists(candidate_path):
paths.append(candidate_path)
else:
paths.append(None)
mask_paths.append(paths)
return image_paths, mask_paths