def split_image(img_path, overwrite=False, cut_dim=CUT_DIMENSION):
""" split two images in single dimension
the input images assume to contain two names in the image name separated by "_"
:param str img_path: path to the input / output image
:param bool overwrite: allow overwrite exiting output images
:param int cut_dim: define splitting dimension
"""
name = os.path.splitext(os.path.basename(img_path))[0]
ext = os.path.splitext(os.path.basename(img_path))[-1]
folder = os.path.dirname(img_path)
obj_names = name.split('_')
paths_img = [
os.path.join(folder, obj_name + ext) for obj_name in obj_names
]
if all(os.path.isfile(p) for p in paths_img) and not overwrite:
logging.debug('existing all splits of %r', paths_img)
return
img = load_large_image(img_path)
# work with just a scaled version
scale_factor = max(1, img.shape[cut_dim] / float(SCALE_SIZE))
sc = 1. / scale_factor
order = cv.INTER_AREA if scale_factor > 1 else cv.INTER_LINEAR
img_small = 255 - cv.resize(img, None, fx=sc, fy=sc, interpolation=order)
img_edge = project_object_edge(img_small, cut_dim)
del img_small
# prepare all cut edges and scale them to original image size
splits = find_split_objects(img_edge, nb_objects=len(obj_names))
if not splits:
logging.error('no splits found for %s', img_path)
return
edges = [
int(round(i * scale_factor)) for i in [0] + splits + [len(img_edge)]
]
# cutting images
for i, path_img_cut in enumerate(paths_img):
if os.path.isfile(path_img_cut) and not overwrite:
logging.debug('existing "%s"', path_img_cut)
continue
if cut_dim == 0:
img_cut = img[edges[i]:edges[i + 1], ...]
elif cut_dim == 1:
img_cut = img[:, edges[i]:edges[i + 1], ...]
else:
raise Exception('unsupported dimension: %i' % cut_dim)
save_large_image(path_img_cut, img_cut)
gc.collect()
time.sleep(1)
def crop_image(img_path, crop_dims=(0, 1), padding=0.15):
""" crop umages to by tight around tissue
:param str img_path: path to image
:param tuple(int) crop_dims: crop in selected dimensions
:param float padding: padding around tissue
"""
img = load_large_image(img_path)
scale_factor = max(1, np.mean(img.shape[:2]) / float(SCALE_SIZE))
# work with just a scaled version
sc = 1. / scale_factor
order = cv.INTER_AREA if scale_factor > 1 else cv.INTER_LINEAR
img_small = 255 - cv.resize(img, None, fx=sc, fy=sc, interpolation=order)
crops = {}
for crop_dim in crop_dims:
if crop_dim not in (0, 1):
raise ValueError('not supported dimension: %i' % crop_dim)
img_edge = project_object_edge(img_small, crop_dim)
begin, end = find_largest_object(img_edge, threshold=TISSUE_CONTENT)
# img_diag = int(np.sqrt(img.shape[0] ** 2 + img.shape[1] ** 2))
pad_px = padding * (end - begin) * scale_factor
begin_px = max(0, int((begin * scale_factor) - pad_px))
end_px = min(img.shape[crop_dim], int((end * scale_factor) + pad_px))
crops[crop_dim] = (begin_px, end_px)
del img_small
for _ in range(2):
if 0 not in crops:
crops[0] = (0, img.shape[0])
img = img[crops[0][0]:crops[0][1], crops[1][0]:crops[1][1], ...]
save_large_image(img_path, img)
gc.collect()
time.sleep(1)