def vessel_enhancement(sample, **kwargs):
"""Apply morphology closing for enhancing vessel structure."""
# size of structural element, chosen to be 11 so that is bigger than the
# largest scale of vessels within dataset
selem_radius = kwargs.get('selem_radius', 11)
@adapt_rgb(each_channel)
def channel_closing(img):
img_closed = morphology.closing(img,
selem=morphology.disk(selem_radius))
return img_closed
def enhancing(img):
img = np.array(img)
img_closed = channel_closing(img)
num_channels = np.shape(img_closed)[-1]
img_closed_channel_mean = np.reshape(
img_closed,
(-1, num_channels)).mean(0).reshape(1, 1, num_channels)
img = img - img_closed + img_closed_channel_mean
return Image.fromarray(img.astype(np.uint8))
sample = image_only_process(enhancing)(sample)
return sample
def random_adjust_saturation(sample, **kwargs):
"""Randomly apply saturation adjustment for RGB image data of a sample."""
saturation_factor_range = kwargs.get('saturation_factor_range', (0.8, 1.2))
saturation_factor = random.uniform(*saturation_factor_range)
sample = image_only_process(
lambda img: TF.adjust_saturation(img, saturation_factor))(sample)
return sample
def random_adjust_gamma(sample, **kwargs):
"""Randomly apply gamma adjustment for RGB image data of a sample."""
gamma_range = kwargs.get('gamma_range', (0.8, 1.2))
gamma = random.uniform(*gamma_range)
sample = image_only_process(lambda img: TF.adjust_gamma(img, gamma))(
sample)
return sample
def random_adjust_contrast(sample, **kwargs):
"""Randomly apply contrast adjustment for RGB image data of a sample."""
contrast_factor_range = kwargs.get('contrast_factor_range', (0.8, 1.2))
contrast_factor = random.uniform(*contrast_factor_range)
sample = image_only_process(
lambda img: TF.adjust_contrast(img, contrast_factor))(sample)
return sample
def random_adjust_brightness(sample, **kwargs):
"""Randomly apply brightness adjustment for RGB image data of a sample."""
brightness_factor_range = kwargs.get('brightness_factor_range', (0.8, 1.2))
brightness_factor = random.uniform(*brightness_factor_range)
sample = image_only_process(
lambda img: TF.adjust_brightness(img, brightness_factor))(sample)
return sample
def normalization(sample, **kwargs):
"""Apply channel wise normalize of image data of a sample."""
channel_mean = kwargs.get('channel_mean', (0.5, 0.5, 0.5))
channel_std = kwargs.get('channel_std', (0.5, 0.5, 0.5))
sample = image_only_process(lambda img: TF.to_pil_image(
TF.normalize(TF.to_tensor(img), channel_mean, channel_std)))(sample)
return sample