qurator-spk/sbb_binarization
1
0
Fork 0
mirror of https://github.com/qurator-spk/sbb_binarization.git synced 2025-06-08 03:40:14 +02:00
Code Issues Releases Wiki Activity

SbbBinarizer: refactor (variable names, less instance-wide state)

Browse source
This commit is contained in:
Konstantin Baierer 2020-10-16 12:21:44 +02:00
parent a1c8f6f465
commit 1fa581283c
3 changed files with 51 additions and 73 deletions
Download patch file Download diff file Expand all files Collapse all files

9
sbb_binarize/cli.py
View file

@ -16,13 +16,8 @@ def main():
options = parser.parse_args() options = parser.parse_args()
binarizer = SbbBinarizer( binarizer = SbbBinarizer(model_dir=options.model)
image_path=options.image, binarizer.run(image_path=options.image, patches=options.patches, save=options.save)
model=options.model,
patches=options.patches,
save=options.save
)
binarizer.run()
if __name__ == "__main__": if __name__ == "__main__":
main() main()

19
sbb_binarize/ocrd_cli.py
View file

@ -39,14 +39,6 @@ class SbbBinarizeProcessor(Processor):
kwargs['version'] = OCRD_TOOL['version'] kwargs['version'] = OCRD_TOOL['version']
super().__init__(*args, **kwargs) super().__init__(*args, **kwargs)
def _run_binarizer(self, img):
return cv2pil(
SbbBinarizer(
image=pil2cv(img),
model=self.model_path,
patches=self.use_patches,
save=None).run())
def process(self): def process(self):
""" """
Binarize with sbb_binarization Binarize with sbb_binarization
@ -56,8 +48,9 @@ class SbbBinarizeProcessor(Processor):
assert_file_grp_cardinality(self.output_file_grp, 1) assert_file_grp_cardinality(self.output_file_grp, 1)
oplevel = self.parameter['operation_level'] oplevel = self.parameter['operation_level']
self.use_patches = self.parameter['patches'] # pylint: disable=attribute-defined-outside-init use_patches = self.parameter['patches'] # pylint: disable=attribute-defined-outside-init
self.model_path = self.parameter['model'] # pylint: disable=attribute-defined-outside-init model_path = self.parameter['model'] # pylint: disable=attribute-defined-outside-init
binarizer = SbbBinarizer(model_dir=self.model_path)
for n, input_file in enumerate(self.input_files): for n, input_file in enumerate(self.input_files):
file_id = make_file_id(input_file, self.output_file_grp) file_id = make_file_id(input_file, self.output_file_grp)
@ -71,7 +64,7 @@ class SbbBinarizeProcessor(Processor):
if oplevel == 'page': if oplevel == 'page':
LOG.info("Binarizing on 'page' level in page '%s'", page_id) LOG.info("Binarizing on 'page' level in page '%s'", page_id)
page_image, page_xywh, _ = self.workspace.image_from_page(page, page_id, feature_filter='binarized') page_image, page_xywh, _ = self.workspace.image_from_page(page, page_id, feature_filter='binarized')
bin_image = self._run_binarizer(page_image) bin_image = cv2pil(binarizer.run(image=pil2cv(page_image), patches=use_patches))
# update METS (add the image file): # update METS (add the image file):
bin_image_path = self.workspace.save_image_file(bin_image, bin_image_path = self.workspace.save_image_file(bin_image,
file_id + '.IMG-BIN', file_id + '.IMG-BIN',
@ -85,7 +78,7 @@ class SbbBinarizeProcessor(Processor):
LOG.warning("Page '%s' contains no text/table regions", page_id) LOG.warning("Page '%s' contains no text/table regions", page_id)
for region in regions: for region in regions:
region_image, region_xywh = self.workspace.image_from_segment(region, page_image, page_xywh, feature_filter='binarized') region_image, region_xywh = self.workspace.image_from_segment(region, page_image, page_xywh, feature_filter='binarized')
region_image_bin = self._run_binarizer(region_image) region_image_bin = cv2pil(binarizer.run(image=pil2cv(region_image), patches=use_patches))
region_image_bin_path = self.workspace.save_image_file( region_image_bin_path = self.workspace.save_image_file(
region_image_bin, region_image_bin,
"%s_%s.IMG-BIN" % (file_id, region.id), "%s_%s.IMG-BIN" % (file_id, region.id),
@ -100,7 +93,7 @@ class SbbBinarizeProcessor(Processor):
LOG.warning("Page '%s' contains no text lines", page_id) LOG.warning("Page '%s' contains no text lines", page_id)
for region_id, line in region_line_tuples: for region_id, line in region_line_tuples:
line_image, line_xywh = self.workspace.image_from_segment(line, page_image, page_xywh, feature_filter='binarized') line_image, line_xywh = self.workspace.image_from_segment(line, page_image, page_xywh, feature_filter='binarized')
line_image_bin = self._run_binarizer(line_image) line_image_bin = cv2pil(binarizer.run(image=pil2cv(line_image), patches=use_patches))
line_image_bin_path = self.workspace.save_image_file( line_image_bin_path = self.workspace.save_image_file(
line_image_bin, line_image_bin,
"%s_%s_%s.IMG-BIN" % (file_id, region_id, line.id), "%s_%s_%s.IMG-BIN" % (file_id, region_id, line.id),

96
sbb_binarize/sbb_binarize.py
View file

@ -22,50 +22,35 @@ def resize_image(img_in, input_height, input_width):
class SbbBinarizer: class SbbBinarizer:
# TODO use True/False for patches def __init__(self, model_dir):
def __init__(self, model, image=None, image_path=None, patches='false', save=None): self.model_dir = model_dir
if (image is not None and image_path is not None) or \
(image is None and image_path is None):
raise ValueError("Must pass either a opencv2 image or an image_path")
if image is not None:
self.image = image
else:
self.image = cv2.imread(self.image)
self.patches = patches
self.save = save
self.model_dir = model
def start_new_session_and_model(self): def start_new_session(self):
config = tf.ConfigProto() config = tf.ConfigProto()
config.gpu_options.allow_growth = True config.gpu_options.allow_growth = True
self.session = tf.Session(config=config) # tf.InteractiveSession() self.session = tf.Session(config=config) # tf.InteractiveSession()
def load_model(self, model_name):
self.model = load_model(join(self.model_dir, model_name), compile=False)
self.img_height = self.model.layers[len(self.model.layers)-1].output_shape[1]
self.img_width = self.model.layers[len(self.model.layers)-1].output_shape[2]
self.n_classes = self.model.layers[len(self.model.layers)-1].output_shape[3]
def end_session(self): def end_session(self):
self.session.close() self.session.close()
del self.model
del self.session del self.session
def predict(self,model_name): def load_model(self, model_name):
self.load_model(model_name) model = load_model(join(self.model_dir, model_name), compile=False)
img = self.image model_height = model.layers[len(model.layers)-1].output_shape[1]
img_width_model = self.img_width model_width = model.layers[len(model.layers)-1].output_shape[2]
img_height_model = self.img_height n_classes = model.layers[len(model.layers)-1].output_shape[3]
return model, model_height, model_width, n_classes
if self.patches in ('true', 'True'): def predict(self, model_name, img, patches):
model, model_height, model_width, n_classes = self.load_model(model_name)
margin = int(0.1 * img_width_model) if patches in ('true', 'True'):
width_mid = img_width_model - 2 * margin margin = int(0.1 * model_width)
height_mid = img_height_model - 2 * margin
width_mid = model_width - 2 * margin
height_mid = model_height - 2 * margin
img = img / float(255.0) img = img / float(255.0)
@ -93,28 +78,28 @@ class SbbBinarizer:
if i == 0: if i == 0:
index_x_d = i * width_mid index_x_d = i * width_mid
index_x_u = index_x_d + img_width_model index_x_u = index_x_d + model_width
elif i > 0: elif i > 0:
index_x_d = i * width_mid index_x_d = i * width_mid
index_x_u = index_x_d + img_width_model index_x_u = index_x_d + model_width
if j == 0: if j == 0:
index_y_d = j * height_mid index_y_d = j * height_mid
index_y_u = index_y_d + img_height_model index_y_u = index_y_d + model_height
elif j > 0: elif j > 0:
index_y_d = j * height_mid index_y_d = j * height_mid
index_y_u = index_y_d + img_height_model index_y_u = index_y_d + model_height
if index_x_u > img_w: if index_x_u > img_w:
index_x_u = img_w index_x_u = img_w
index_x_d = img_w - img_width_model index_x_d = img_w - model_width
if index_y_u > img_h: if index_y_u > img_h:
index_y_u = img_h index_y_u = img_h
index_y_d = img_h - img_height_model index_y_d = img_h - model_height
img_patch = img[index_y_d:index_y_u, index_x_d:index_x_u, :] img_patch = img[index_y_d:index_y_u, index_x_d:index_x_u, :]
label_p_pred = self.model.predict(img_patch.reshape(1, img_patch.shape[0], img_patch.shape[1], img_patch.shape[2])) label_p_pred = model.predict(img_patch.reshape(1, img_patch.shape[0], img_patch.shape[1], img_patch.shape[2]))
seg = np.argmax(label_p_pred, axis=3)[0] seg = np.argmax(label_p_pred, axis=3)[0]
@ -189,10 +174,9 @@ class SbbBinarizer:
img_h_page = img.shape[0] img_h_page = img.shape[0]
img_w_page = img.shape[1] img_w_page = img.shape[1]
img = img / float(255.0) img = img / float(255.0)
img = resize_image(img, img_height_model, img_width_model) img = resize_image(img, model_height, model_width)
label_p_pred = self.model.predict( label_p_pred = model.predict(img.reshape(1, img.shape[0], img.shape[1], img.shape[2]))
img.reshape(1, img.shape[0], img.shape[1], img.shape[2]))
seg = np.argmax(label_p_pred, axis=3)[0] seg = np.argmax(label_p_pred, axis=3)[0]
seg_color = np.repeat(seg[:, :, np.newaxis], 3, axis=2) seg_color = np.repeat(seg[:, :, np.newaxis], 3, axis=2)
@ -200,29 +184,35 @@ class SbbBinarizer:
prediction_true = prediction_true.astype(np.uint8) prediction_true = prediction_true.astype(np.uint8)
return prediction_true[:,:,0] return prediction_true[:,:,0]
def run(self): # TODO use True/False for patches
self.start_new_session_and_model() def run(self, image=None, image_path=None, save=None, patches='false'):
models_n = listdir(self.model_dir) if (image is not None and image_path is not None) or \
(image is None and image_path is None):
raise ValueError("Must pass either a opencv2 image or an image_path")
if image_path is not None:
image = cv2.imread(image)
self.start_new_session()
list_of_model_files = listdir(self.model_dir)
img_last = 0 img_last = 0
for model_in in models_n: for model_in in list_of_model_files:
res = self.predict(model_in) res = self.predict(model_in, image, patches)
img_fin = np.zeros((res.shape[0], res.shape[1], 3)) img_fin = np.zeros((res.shape[0], res.shape[1], 3))
res[:, :][res[:, :] == 0] = 2 res[:, :][res[:, :] == 0] = 2
res = res-1 res = res - 1
res = res*255 res = res * 255
img_fin[:, :, 0] = res img_fin[:, :, 0] = res
img_fin[:, :, 1] = res img_fin[:, :, 1] = res
img_fin[:, :, 2] = res img_fin[:, :, 2] = res
img_fin = img_fin.astype(np.uint8) img_fin = img_fin.astype(np.uint8)
img_fin = (res[:, :] == 0)*255 img_fin = (res[:, :] == 0) * 255
img_last = img_last+img_fin img_last = img_last + img_fin
kernel = np.ones((5, 5), np.uint8) kernel = np.ones((5, 5), np.uint8)
img_last[:, :][img_last[:, :] > 0] = 255 img_last[:, :][img_last[:, :] > 0] = 255
img_last = (img_last[:, :] == 0)*255 img_last = (img_last[:, :] == 0) * 255
if self.save: if save:
cv2.imwrite(self.save, img_last) cv2.imwrite(save, img_last)
return img_last return img_last