qurator-spk/eynollah
1
0
Fork 0
mirror of https://github.com/qurator-spk/eynollah.git synced 2025-06-07 19:35:01 +02:00
Code Issues Projects Releases Wiki Activity

Merge branch 'refactor' into main

Browse source
This commit is contained in:
Konstantin Baierer 2020-12-16 12:51:20 +01:00
commit 044ff0c5a2
19 changed files with 10709 additions and 11737 deletions
Download patch file Download diff file Expand all files Collapse all files

28
.circleci/config.yml Normal file
View file

@ -0,0 +1,28 @@
version: 2
jobs:
build-python36:
docker:
- image: python:3.6
steps:
- checkout
- restore_cache:
keys:
- model-cache
- run: make models
- save_cache:
key: model-cache
paths:
models_eynollah.tar.gz
models_eynollah
- run: make install
- run: make smoke-test
workflows:
version: 2
build:
jobs:
- build-python36
#- build-python37
#- build-python38 # no tensorflow for python 3.8

3
.gitignore vendored
View file

@ -1,2 +1,5 @@
*.egg-info
__pycache__
sbb_newspapers_org_image/pylint.log
models_eynollah*
output.html

19
Makefile
View file

@ -4,14 +4,26 @@ help:
@echo ""
@echo " Targets"
@echo ""
@echo " models Download and extract models to $(PWD)/models_eynollah"
@echo " install Install with pip"
@echo " install-dev Install editable with pip"
@echo " test Run unit tests"
@echo ""
@echo " Variables"
@echo ""
# END-EVAL
# Download and extract models to $(PWD)/models_eynollah
models: models_eynollah
models_eynollah: models_eynollah.tar.gz
tar xf models_eynollah.tar.gz
models_eynollah.tar.gz:
wget 'https://qurator-data.de/eynollah/models_eynollah.tar.gz'
# Install with pip
install:
pip install .
@ -19,3 +31,10 @@ install:
# Install editable with pip
install-dev:
pip install -e .
smoke-test:
eynollah -i tests/resources/kant_aufklaerung_1784_0020.tif -o . -m $(PWD)/models_eynollah
# Run unit tests
test:
pytest tests

2
requirements-test.txt Normal file
View file

@ -0,0 +1,2 @@
pytest
black

4
requirements.txt
View file

@ -1,7 +1,7 @@
# ocrd includes opencv, numpy, shapely, click
ocrd >= 2.20.1
seaborn >= 0.11.0
keras >= 2.3.1
scikit-learn >= 0.23.2
tensorflow >= 1.15, < 2
tensorflow-gpu >= 1.15, < 2
imutils >= 0.5.3
matplotlib

107
sbb_newspapers_org_image/cli.py Normal file
View file

@ -0,0 +1,107 @@
import click
from sbb_newspapers_org_image.eynollah import eynollah
@click.command()
@click.option(
"--image", "-i", help="image filename", type=click.Path(exists=True, dir_okay=False)
)
@click.option(
"--out",
"-o",
help="directory to write output xml data",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--model",
"-m",
help="directory of models",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--save_images",
"-si",
help="if a directory is given, images in documents will be cropped and saved there",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--save_layout",
"-sl",
help="if a directory is given, plot of layout will be saved there",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--save_deskewed",
"-sd",
help="if a directory is given, deskewed image will be saved there",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--save_all",
"-sa",
help="if a directory is given, all plots needed for documentation will be saved there",
type=click.Path(exists=True, file_okay=False),
)
@click.option(
"--allow_enhancement",
"-ae",
is_flag=True,
help="if this parameter set to true, this tool would check that input image need resizing and enhancement or not. If so output of resized and enhanced image and corresponding layout data will be written in out directory",
)
@click.option(
"--curved_line",
"-cl",
is_flag=True,
help="if this parameter set to true, this tool will try to return contoure of textlines instead of rectabgle bounding box of textline. This should be taken into account that with this option the tool need more time to do process.",
)
@click.option(
"--full_layout",
"-fl",
is_flag=True,
help="if this parameter set to true, this tool will try to return all elements of layout.",
)
@click.option(
"--allow_scaling",
"-as",
is_flag=True,
help="if this parameter set to true, this tool would check the scale and if needed it will scale it to perform better layout detection",
)
@click.option(
"--headers_off",
"-ho",
is_flag=True,
help="if this parameter set to true, this tool would ignore headers role in reading order",
)
def main(
image,
out,
model,
save_images,
save_layout,
save_deskewed,
save_all,
allow_enhancement,
curved_line,
full_layout,
allow_scaling,
headers_off,
):
eynollah(
image,
None,
out,
model,
save_images,
save_layout,
save_deskewed,
save_all,
allow_enhancement,
curved_line,
full_layout,
allow_scaling,
headers_off,
).run()
if __name__ == "__main__":
main()

12751
sbb_newspapers_org_image/eynollah.py
View file

File diff suppressed because it is too large Load diff

3029
sbb_newspapers_org_image/unused.py Normal file
View file

File diff suppressed because it is too large Load diff

2091
sbb_newspapers_org_image/utils/__init__.py Normal file
View file

File diff suppressed because it is too large Load diff

298
sbb_newspapers_org_image/utils/contour.py Normal file
View file

@ -0,0 +1,298 @@
import cv2
import numpy as np
from shapely import geometry
from .rotate import rotate_image, rotation_image_new
def contours_in_same_horizon(cy_main_hor):
X1 = np.zeros((len(cy_main_hor), len(cy_main_hor)))
X2 = np.zeros((len(cy_main_hor), len(cy_main_hor)))
X1[0::1, :] = cy_main_hor[:]
X2 = X1.T
X_dif = np.abs(X2 - X1)
args_help = np.array(range(len(cy_main_hor)))
all_args = []
for i in range(len(cy_main_hor)):
list_h = list(args_help[X_dif[i, :] <= 20])
list_h.append(i)
if len(list_h) > 1:
all_args.append(list(set(list_h)))
return np.unique(all_args)
def find_contours_mean_y_diff(contours_main):
M_main = [cv2.moments(contours_main[j]) for j in range(len(contours_main))]
cy_main = [(M_main[j]["m01"] / (M_main[j]["m00"] + 1e-32)) for j in range(len(M_main))]
return np.mean(np.diff(np.sort(np.array(cy_main))))
def find_features_of_contours(contours_main):
areas_main = np.array([cv2.contourArea(contours_main[j]) for j in range(len(contours_main))])
M_main = [cv2.moments(contours_main[j]) for j in range(len(contours_main))]
cx_main = [(M_main[j]["m10"] / (M_main[j]["m00"] + 1e-32)) for j in range(len(M_main))]
cy_main = [(M_main[j]["m01"] / (M_main[j]["m00"] + 1e-32)) for j in range(len(M_main))]
x_min_main = np.array([np.min(contours_main[j][:, 0, 0]) for j in range(len(contours_main))])
x_max_main = np.array([np.max(contours_main[j][:, 0, 0]) for j in range(len(contours_main))])
y_min_main = np.array([np.min(contours_main[j][:, 0, 1]) for j in range(len(contours_main))])
y_max_main = np.array([np.max(contours_main[j][:, 0, 1]) for j in range(len(contours_main))])
return y_min_main, y_max_main, areas_main
def return_contours_of_interested_region_and_bounding_box(region_pre_p, pixel):
# pixels of images are identified by 5
cnts_images = (region_pre_p[:, :, 0] == pixel) * 1
cnts_images = cnts_images.astype(np.uint8)
cnts_images = np.repeat(cnts_images[:, :, np.newaxis], 3, axis=2)
imgray = cv2.cvtColor(cnts_images, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_imgs, hiearchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contours_imgs = return_parent_contours(contours_imgs, hiearchy)
contours_imgs = filter_contours_area_of_image_tables(thresh, contours_imgs, hiearchy, max_area=1, min_area=0.0003)
boxes = []
for jj in range(len(contours_imgs)):
x, y, w, h = cv2.boundingRect(contours_imgs[jj])
boxes.append([int(x), int(y), int(w), int(h)])
return contours_imgs, boxes
def get_text_region_boxes_by_given_contours(contours):
kernel = np.ones((5, 5), np.uint8)
boxes = []
contours_new = []
for jj in range(len(contours)):
x, y, w, h = cv2.boundingRect(contours[jj])
boxes.append([x, y, w, h])
contours_new.append(contours[jj])
del contours
return boxes, contours_new
def filter_contours_area_of_image(image, contours, hirarchy, max_area, min_area):
found_polygons_early = list()
jv = 0
for c in contours:
if len(c) < 3: # A polygon cannot have less than 3 points
continue
polygon = geometry.Polygon([point[0] for point in c])
area = polygon.area
if area >= min_area * np.prod(image.shape[:2]) and area <= max_area * np.prod(image.shape[:2]) and hirarchy[0][jv][3] == -1: # and hirarchy[0][jv][3]==-1 :
found_polygons_early.append(np.array([[point] for point in polygon.exterior.coords], dtype=np.uint))
jv += 1
return found_polygons_early
def filter_contours_area_of_image_interiors(image, contours, hirarchy, max_area, min_area):
found_polygons_early = list()
jv = 0
for c in contours:
if len(c) < 3: # A polygon cannot have less than 3 points
continue
polygon = geometry.Polygon([point[0] for point in c])
area = polygon.area
if area >= min_area * np.prod(image.shape[:2]) and area <= max_area * np.prod(image.shape[:2]) and hirarchy[0][jv][3] != -1:
# print(c[0][0][1])
found_polygons_early.append(np.array([point for point in polygon.exterior.coords], dtype=np.uint))
jv += 1
return found_polygons_early
def filter_contours_area_of_image_tables(image, contours, hirarchy, max_area, min_area):
found_polygons_early = list()
jv = 0
for c in contours:
if len(c) < 3: # A polygon cannot have less than 3 points
continue
polygon = geometry.Polygon([point[0] for point in c])
# area = cv2.contourArea(c)
area = polygon.area
##print(np.prod(thresh.shape[:2]))
# Check that polygon has area greater than minimal area
# print(hirarchy[0][jv][3],hirarchy )
if area >= min_area * np.prod(image.shape[:2]) and area <= max_area * np.prod(image.shape[:2]): # and hirarchy[0][jv][3]==-1 :
# print(c[0][0][1])
found_polygons_early.append(np.array([[point] for point in polygon.exterior.coords], dtype=np.int32))
jv += 1
return found_polygons_early
def find_new_features_of_contoures(contours_main):
areas_main = np.array([cv2.contourArea(contours_main[j]) for j in range(len(contours_main))])
M_main = [cv2.moments(contours_main[j]) for j in range(len(contours_main))]
cx_main = [(M_main[j]["m10"] / (M_main[j]["m00"] + 1e-32)) for j in range(len(M_main))]
cy_main = [(M_main[j]["m01"] / (M_main[j]["m00"] + 1e-32)) for j in range(len(M_main))]
try:
x_min_main = np.array([np.min(contours_main[j][:, 0, 0]) for j in range(len(contours_main))])
argmin_x_main = np.array([np.argmin(contours_main[j][:, 0, 0]) for j in range(len(contours_main))])
x_min_from_argmin = np.array([contours_main[j][argmin_x_main[j], 0, 0] for j in range(len(contours_main))])
y_corr_x_min_from_argmin = np.array([contours_main[j][argmin_x_main[j], 0, 1] for j in range(len(contours_main))])
x_max_main = np.array([np.max(contours_main[j][:, 0, 0]) for j in range(len(contours_main))])
y_min_main = np.array([np.min(contours_main[j][:, 0, 1]) for j in range(len(contours_main))])
y_max_main = np.array([np.max(contours_main[j][:, 0, 1]) for j in range(len(contours_main))])
except:
x_min_main = np.array([np.min(contours_main[j][:, 0]) for j in range(len(contours_main))])
argmin_x_main = np.array([np.argmin(contours_main[j][:, 0]) for j in range(len(contours_main))])
x_min_from_argmin = np.array([contours_main[j][argmin_x_main[j], 0] for j in range(len(contours_main))])
y_corr_x_min_from_argmin = np.array([contours_main[j][argmin_x_main[j], 1] for j in range(len(contours_main))])
x_max_main = np.array([np.max(contours_main[j][:, 0]) for j in range(len(contours_main))])
y_min_main = np.array([np.min(contours_main[j][:, 1]) for j in range(len(contours_main))])
y_max_main = np.array([np.max(contours_main[j][:, 1]) for j in range(len(contours_main))])
# dis_x=np.abs(x_max_main-x_min_main)
return cx_main, cy_main, x_min_main, x_max_main, y_min_main, y_max_main, y_corr_x_min_from_argmin
def return_parent_contours(contours, hierarchy):
contours_parent = [contours[i] for i in range(len(contours)) if hierarchy[0][i][3] == -1]
return contours_parent
def return_contours_of_interested_region(region_pre_p, pixel, min_area=0.0002):
# pixels of images are identified by 5
if len(region_pre_p.shape) == 3:
cnts_images = (region_pre_p[:, :, 0] == pixel) * 1
else:
cnts_images = (region_pre_p[:, :] == pixel) * 1
cnts_images = cnts_images.astype(np.uint8)
cnts_images = np.repeat(cnts_images[:, :, np.newaxis], 3, axis=2)
imgray = cv2.cvtColor(cnts_images, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_imgs, hiearchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contours_imgs = return_parent_contours(contours_imgs, hiearchy)
contours_imgs = filter_contours_area_of_image_tables(thresh, contours_imgs, hiearchy, max_area=1, min_area=min_area)
return contours_imgs
def get_textregion_contours_in_org_image(cnts, img, slope_first):
cnts_org = []
# print(cnts,'cnts')
for i in range(len(cnts)):
img_copy = np.zeros(img.shape)
img_copy = cv2.fillPoly(img_copy, pts=[cnts[i]], color=(1, 1, 1))
# plt.imshow(img_copy)
# plt.show()
# print(img.shape,'img')
img_copy = rotation_image_new(img_copy, -slope_first)
##print(img_copy.shape,'img_copy')
# plt.imshow(img_copy)
# plt.show()
img_copy = img_copy.astype(np.uint8)
imgray = cv2.cvtColor(img_copy, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
cont_int, _ = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
cont_int[0][:, 0, 0] = cont_int[0][:, 0, 0] + np.abs(img_copy.shape[1] - img.shape[1])
cont_int[0][:, 0, 1] = cont_int[0][:, 0, 1] + np.abs(img_copy.shape[0] - img.shape[0])
# print(np.shape(cont_int[0]))
cnts_org.append(cont_int[0])
# print(cnts_org,'cnts_org')
# sys.exit()
# self.y_shift = np.abs(img_copy.shape[0] - img.shape[0])
# self.x_shift = np.abs(img_copy.shape[1] - img.shape[1])
return cnts_org
def return_contours_of_interested_textline(region_pre_p, pixel):
# pixels of images are identified by 5
if len(region_pre_p.shape) == 3:
cnts_images = (region_pre_p[:, :, 0] == pixel) * 1
else:
cnts_images = (region_pre_p[:, :] == pixel) * 1
cnts_images = cnts_images.astype(np.uint8)
cnts_images = np.repeat(cnts_images[:, :, np.newaxis], 3, axis=2)
imgray = cv2.cvtColor(cnts_images, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_imgs, hiearchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contours_imgs = return_parent_contours(contours_imgs, hiearchy)
contours_imgs = filter_contours_area_of_image_tables(thresh, contours_imgs, hiearchy, max_area=1, min_area=0.000000003)
return contours_imgs
def return_bonding_box_of_contours(cnts):
boxes_tot = []
for i in range(len(cnts)):
x, y, w, h = cv2.boundingRect(cnts[i])
box = [x, y, w, h]
boxes_tot.append(box)
return boxes_tot
def return_contours_of_image(image):
if len(image.shape) == 2:
image = np.repeat(image[:, :, np.newaxis], 3, axis=2)
image = image.astype(np.uint8)
else:
image = image.astype(np.uint8)
imgray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours, hierachy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
return contours, hierachy
def return_contours_of_interested_region_by_min_size(region_pre_p, pixel, min_size=0.00003):
# pixels of images are identified by 5
if len(region_pre_p.shape) == 3:
cnts_images = (region_pre_p[:, :, 0] == pixel) * 1
else:
cnts_images = (region_pre_p[:, :] == pixel) * 1
cnts_images = cnts_images.astype(np.uint8)
cnts_images = np.repeat(cnts_images[:, :, np.newaxis], 3, axis=2)
imgray = cv2.cvtColor(cnts_images, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_imgs, hiearchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contours_imgs = return_parent_contours(contours_imgs, hiearchy)
contours_imgs = filter_contours_area_of_image_tables(thresh, contours_imgs, hiearchy, max_area=1, min_area=min_size)
return contours_imgs
def return_contours_of_interested_region_by_size(region_pre_p, pixel, min_area, max_area):
# pixels of images are identified by 5
if len(region_pre_p.shape) == 3:
cnts_images = (region_pre_p[:, :, 0] == pixel) * 1
else:
cnts_images = (region_pre_p[:, :] == pixel) * 1
cnts_images = cnts_images.astype(np.uint8)
cnts_images = np.repeat(cnts_images[:, :, np.newaxis], 3, axis=2)
imgray = cv2.cvtColor(cnts_images, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_imgs, hiearchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
contours_imgs = return_parent_contours(contours_imgs, hiearchy)
contours_imgs = filter_contours_area_of_image_tables(thresh, contours_imgs, hiearchy, max_area=max_area, min_area=min_area)
img_ret = np.zeros((region_pre_p.shape[0], region_pre_p.shape[1], 3))
img_ret = cv2.fillPoly(img_ret, pts=contours_imgs, color=(1, 1, 1))
return img_ret[:, :, 0]

501
sbb_newspapers_org_image/utils/drop_capitals.py Normal file
View file

@ -0,0 +1,501 @@
import numpy as np
import cv2
from .contour import (
find_new_features_of_contoures,
return_contours_of_image,
return_parent_contours,
)
def adhere_drop_capital_region_into_cprresponding_textline(
text_regions_p,
polygons_of_drop_capitals,
contours_only_text_parent,
contours_only_text_parent_h,
all_box_coord,
all_box_coord_h,
all_found_texline_polygons,
all_found_texline_polygons_h,
kernel=None,
curved_line=False,
):
# print(np.shape(all_found_texline_polygons),np.shape(all_found_texline_polygons[3]),'all_found_texline_polygonsshape')
# print(all_found_texline_polygons[3])
cx_m, cy_m, _, _, _, _, _ = find_new_features_of_contoures(contours_only_text_parent)
cx_h, cy_h, _, _, _, _, _ = find_new_features_of_contoures(contours_only_text_parent_h)
cx_d, cy_d, _, _, y_min_d, y_max_d, _ = find_new_features_of_contoures(polygons_of_drop_capitals)
img_con_all = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
for j_cont in range(len(contours_only_text_parent)):
img_con_all[all_box_coord[j_cont][0] : all_box_coord[j_cont][1], all_box_coord[j_cont][2] : all_box_coord[j_cont][3], 0] = (j_cont + 1) * 3
# img_con_all=cv2.fillPoly(img_con_all,pts=[contours_only_text_parent[j_cont]],color=((j_cont+1)*3,(j_cont+1)*3,(j_cont+1)*3))
# plt.imshow(img_con_all[:,:,0])
# plt.show()
# img_con_all=cv2.dilate(img_con_all, kernel, iterations=3)
# plt.imshow(img_con_all[:,:,0])
# plt.show()
# print(np.unique(img_con_all[:,:,0]))
for i_drop in range(len(polygons_of_drop_capitals)):
# print(i_drop,'i_drop')
img_con_all_copy = np.copy(img_con_all)
img_con = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
img_con = cv2.fillPoly(img_con, pts=[polygons_of_drop_capitals[i_drop]], color=(1, 1, 1))
# plt.imshow(img_con[:,:,0])
# plt.show()
##img_con=cv2.dilate(img_con, kernel, iterations=30)
# plt.imshow(img_con[:,:,0])
# plt.show()
# print(np.unique(img_con[:,:,0]))
img_con_all_copy[:, :, 0] = img_con_all_copy[:, :, 0] + img_con[:, :, 0]
img_con_all_copy[:, :, 0][img_con_all_copy[:, :, 0] == 1] = 0
kherej_ghesmat = np.unique(img_con_all_copy[:, :, 0]) / 3
res_summed_pixels = np.unique(img_con_all_copy[:, :, 0]) % 3
region_with_intersected_drop = kherej_ghesmat[res_summed_pixels == 1]
# region_with_intersected_drop=region_with_intersected_drop/3
region_with_intersected_drop = region_with_intersected_drop.astype(np.uint8)
# print(len(region_with_intersected_drop),'region_with_intersected_drop1')
if len(region_with_intersected_drop) == 0:
img_con_all_copy = np.copy(img_con_all)
img_con = cv2.dilate(img_con, kernel, iterations=4)
img_con_all_copy[:, :, 0] = img_con_all_copy[:, :, 0] + img_con[:, :, 0]
img_con_all_copy[:, :, 0][img_con_all_copy[:, :, 0] == 1] = 0
kherej_ghesmat = np.unique(img_con_all_copy[:, :, 0]) / 3
res_summed_pixels = np.unique(img_con_all_copy[:, :, 0]) % 3
region_with_intersected_drop = kherej_ghesmat[res_summed_pixels == 1]
# region_with_intersected_drop=region_with_intersected_drop/3
region_with_intersected_drop = region_with_intersected_drop.astype(np.uint8)
# print(np.unique(img_con_all_copy[:,:,0]))
if curved_line:
if len(region_with_intersected_drop) > 1:
sum_pixels_of_intersection = []
for i in range(len(region_with_intersected_drop)):
# print((region_with_intersected_drop[i]*3+1))
sum_pixels_of_intersection.append(((img_con_all_copy[:, :, 0] == (region_with_intersected_drop[i] * 3 + 1)) * 1).sum())
# print(sum_pixels_of_intersection)
region_final = region_with_intersected_drop[np.argmax(sum_pixels_of_intersection)] - 1
# print(region_final,'region_final')
# cx_t,cy_t ,_, _, _ ,_,_= find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
try:
cx_t, cy_t, _, _, _, _, _ = find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
# print(all_box_coord[j_cont])
# print(cx_t)
# print(cy_t)
# print(cx_d[i_drop])
# print(cy_d[i_drop])
y_lines = np.array(cy_t) # all_box_coord[int(region_final)][0]+np.array(cy_t)
# print(y_lines)
y_lines[y_lines < y_min_d[i_drop]] = 0
# print(y_lines)
arg_min = np.argmin(np.abs(y_lines - y_min_d[i_drop]))
# print(arg_min)
cnt_nearest = np.copy(all_found_texline_polygons[int(region_final)][arg_min])
cnt_nearest[:, 0, 0] = all_found_texline_polygons[int(region_final)][arg_min][:, 0, 0] # +all_box_coord[int(region_final)][2]
cnt_nearest[:, 0, 1] = all_found_texline_polygons[int(region_final)][arg_min][:, 0, 1] # +all_box_coord[int(region_final)][0]
img_textlines = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
img_textlines = cv2.fillPoly(img_textlines, pts=[cnt_nearest], color=(255, 255, 255))
img_textlines = cv2.fillPoly(img_textlines, pts=[polygons_of_drop_capitals[i_drop]], color=(255, 255, 255))
img_textlines = img_textlines.astype(np.uint8)
imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_combined, hierachy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# print(len(contours_combined),'len textlines mixed')
areas_cnt_text = np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
contours_biggest = contours_combined[np.argmax(areas_cnt_text)]
# print(np.shape(contours_biggest))
# print(contours_biggest[:])
# contours_biggest[:,0,0]=contours_biggest[:,0,0]#-all_box_coord[int(region_final)][2]
# contours_biggest[:,0,1]=contours_biggest[:,0,1]#-all_box_coord[int(region_final)][0]
# contours_biggest=contours_biggest.reshape(np.shape(contours_biggest)[0],np.shape(contours_biggest)[2])
all_found_texline_polygons[int(region_final)][arg_min] = contours_biggest
except:
# print('gordun1')
pass
elif len(region_with_intersected_drop) == 1:
region_final = region_with_intersected_drop[0] - 1
# areas_main=np.array([cv2.contourArea(all_found_texline_polygons[int(region_final)][0][j] ) for j in range(len(all_found_texline_polygons[int(region_final)]))])
# cx_t,cy_t ,_, _, _ ,_,_= find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
cx_t, cy_t, _, _, _, _, _ = find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
# print(all_box_coord[j_cont])
# print(cx_t)
# print(cy_t)
# print(cx_d[i_drop])
# print(cy_d[i_drop])
y_lines = np.array(cy_t) # all_box_coord[int(region_final)][0]+np.array(cy_t)
y_lines[y_lines < y_min_d[i_drop]] = 0
# print(y_lines)
arg_min = np.argmin(np.abs(y_lines - y_min_d[i_drop]))
# print(arg_min)
cnt_nearest = np.copy(all_found_texline_polygons[int(region_final)][arg_min])
cnt_nearest[:, 0, 0] = all_found_texline_polygons[int(region_final)][arg_min][:, 0, 0] # +all_box_coord[int(region_final)][2]
cnt_nearest[:, 0, 1] = all_found_texline_polygons[int(region_final)][arg_min][:, 0, 1] # +all_box_coord[int(region_final)][0]
img_textlines = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
img_textlines = cv2.fillPoly(img_textlines, pts=[cnt_nearest], color=(255, 255, 255))
img_textlines = cv2.fillPoly(img_textlines, pts=[polygons_of_drop_capitals[i_drop]], color=(255, 255, 255))
img_textlines = img_textlines.astype(np.uint8)
# plt.imshow(img_textlines)
# plt.show()
imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_combined, hierachy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# print(len(contours_combined),'len textlines mixed')
areas_cnt_text = np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
contours_biggest = contours_combined[np.argmax(areas_cnt_text)]
# print(np.shape(contours_biggest))
# print(contours_biggest[:])
# contours_biggest[:,0,0]=contours_biggest[:,0,0]#-all_box_coord[int(region_final)][2]
# contours_biggest[:,0,1]=contours_biggest[:,0,1]#-all_box_coord[int(region_final)][0]
# print(np.shape(contours_biggest),'contours_biggest')
# print(np.shape(all_found_texline_polygons[int(region_final)][arg_min]))
##contours_biggest=contours_biggest.reshape(np.shape(contours_biggest)[0],np.shape(contours_biggest)[2])
all_found_texline_polygons[int(region_final)][arg_min] = contours_biggest
# print(cx_t,'print')
try:
# print(all_found_texline_polygons[j_cont][0])
cx_t, cy_t, _, _, _, _, _ = find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
# print(all_box_coord[j_cont])
# print(cx_t)
# print(cy_t)
# print(cx_d[i_drop])
# print(cy_d[i_drop])
y_lines = all_box_coord[int(region_final)][0] + np.array(cy_t)
y_lines[y_lines < y_min_d[i_drop]] = 0
# print(y_lines)
arg_min = np.argmin(np.abs(y_lines - y_min_d[i_drop]))
# print(arg_min)
cnt_nearest = np.copy(all_found_texline_polygons[int(region_final)][arg_min])
cnt_nearest[:, 0, 0] = all_found_texline_polygons[int(region_final)][arg_min][:, 0, 0] # +all_box_coord[int(region_final)][2]
cnt_nearest[:, 0, 1] = all_found_texline_polygons[int(region_final)][arg_min][:, 0, 1] # +all_box_coord[int(region_final)][0]
img_textlines = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
img_textlines = cv2.fillPoly(img_textlines, pts=[cnt_nearest], color=(255, 255, 255))
img_textlines = cv2.fillPoly(img_textlines, pts=[polygons_of_drop_capitals[i_drop]], color=(255, 255, 255))
img_textlines = img_textlines.astype(np.uint8)
imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_combined, hierachy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# print(len(contours_combined),'len textlines mixed')
areas_cnt_text = np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
contours_biggest = contours_combined[np.argmax(areas_cnt_text)]
# print(np.shape(contours_biggest))
# print(contours_biggest[:])
contours_biggest[:, 0, 0] = contours_biggest[:, 0, 0] # -all_box_coord[int(region_final)][2]
contours_biggest[:, 0, 1] = contours_biggest[:, 0, 1] # -all_box_coord[int(region_final)][0]
##contours_biggest=contours_biggest.reshape(np.shape(contours_biggest)[0],np.shape(contours_biggest)[2])
all_found_texline_polygons[int(region_final)][arg_min] = contours_biggest
# all_found_texline_polygons[int(region_final)][arg_min]=contours_biggest
except:
pass
else:
pass
##cx_t,cy_t ,_, _, _ ,_,_= find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
###print(all_box_coord[j_cont])
###print(cx_t)
###print(cy_t)
###print(cx_d[i_drop])
###print(cy_d[i_drop])
##y_lines=all_box_coord[int(region_final)][0]+np.array(cy_t)
##y_lines[y_lines<y_min_d[i_drop]]=0
###print(y_lines)
##arg_min=np.argmin(np.abs(y_lines-y_min_d[i_drop]) )
###print(arg_min)
##cnt_nearest=np.copy(all_found_texline_polygons[int(region_final)][arg_min])
##cnt_nearest[:,0,0]=all_found_texline_polygons[int(region_final)][arg_min][:,0,0]#+all_box_coord[int(region_final)][2]
##cnt_nearest[:,0,1]=all_found_texline_polygons[int(region_final)][arg_min][:,0,1]#+all_box_coord[int(region_final)][0]
##img_textlines=np.zeros((text_regions_p.shape[0],text_regions_p.shape[1],3))
##img_textlines=cv2.fillPoly(img_textlines,pts=[cnt_nearest],color=(255,255,255))
##img_textlines=cv2.fillPoly(img_textlines,pts=[polygons_of_drop_capitals[i_drop] ],color=(255,255,255))
##img_textlines=img_textlines.astype(np.uint8)
##plt.imshow(img_textlines)
##plt.show()
##imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
##ret, thresh = cv2.threshold(imgray, 0, 255, 0)
##contours_combined,hierachy=cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
##print(len(contours_combined),'len textlines mixed')
##areas_cnt_text=np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
##contours_biggest=contours_combined[np.argmax(areas_cnt_text)]
###print(np.shape(contours_biggest))
###print(contours_biggest[:])
##contours_biggest[:,0,0]=contours_biggest[:,0,0]#-all_box_coord[int(region_final)][2]
##contours_biggest[:,0,1]=contours_biggest[:,0,1]#-all_box_coord[int(region_final)][0]
##contours_biggest=contours_biggest.reshape(np.shape(contours_biggest)[0],np.shape(contours_biggest)[2])
##all_found_texline_polygons[int(region_final)][arg_min]=contours_biggest
else:
if len(region_with_intersected_drop) > 1:
sum_pixels_of_intersection = []
for i in range(len(region_with_intersected_drop)):
# print((region_with_intersected_drop[i]*3+1))
sum_pixels_of_intersection.append(((img_con_all_copy[:, :, 0] == (region_with_intersected_drop[i] * 3 + 1)) * 1).sum())
# print(sum_pixels_of_intersection)
region_final = region_with_intersected_drop[np.argmax(sum_pixels_of_intersection)] - 1
# print(region_final,'region_final')
# cx_t,cy_t ,_, _, _ ,_,_= find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
try:
cx_t, cy_t, _, _, _, _, _ = find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
# print(all_box_coord[j_cont])
# print(cx_t)
# print(cy_t)
# print(cx_d[i_drop])
# print(cy_d[i_drop])
y_lines = all_box_coord[int(region_final)][0] + np.array(cy_t)
# print(y_lines)
y_lines[y_lines < y_min_d[i_drop]] = 0
# print(y_lines)
arg_min = np.argmin(np.abs(y_lines - y_min_d[i_drop]))
# print(arg_min)
cnt_nearest = np.copy(all_found_texline_polygons[int(region_final)][arg_min])
cnt_nearest[:, 0] = all_found_texline_polygons[int(region_final)][arg_min][:, 0] + all_box_coord[int(region_final)][2]
cnt_nearest[:, 1] = all_found_texline_polygons[int(region_final)][arg_min][:, 1] + all_box_coord[int(region_final)][0]
img_textlines = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
img_textlines = cv2.fillPoly(img_textlines, pts=[cnt_nearest], color=(255, 255, 255))
img_textlines = cv2.fillPoly(img_textlines, pts=[polygons_of_drop_capitals[i_drop]], color=(255, 255, 255))
img_textlines = img_textlines.astype(np.uint8)
imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_combined, hierachy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# print(len(contours_combined),'len textlines mixed')
areas_cnt_text = np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
contours_biggest = contours_combined[np.argmax(areas_cnt_text)]
# print(np.shape(contours_biggest))
# print(contours_biggest[:])
contours_biggest[:, 0, 0] = contours_biggest[:, 0, 0] - all_box_coord[int(region_final)][2]
contours_biggest[:, 0, 1] = contours_biggest[:, 0, 1] - all_box_coord[int(region_final)][0]
contours_biggest = contours_biggest.reshape(np.shape(contours_biggest)[0], np.shape(contours_biggest)[2])
all_found_texline_polygons[int(region_final)][arg_min] = contours_biggest
except:
# print('gordun1')
pass
elif len(region_with_intersected_drop) == 1:
region_final = region_with_intersected_drop[0] - 1
# areas_main=np.array([cv2.contourArea(all_found_texline_polygons[int(region_final)][0][j] ) for j in range(len(all_found_texline_polygons[int(region_final)]))])
# cx_t,cy_t ,_, _, _ ,_,_= find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
# print(cx_t,'print')
try:
# print(all_found_texline_polygons[j_cont][0])
cx_t, cy_t, _, _, _, _, _ = find_new_features_of_contoures(all_found_texline_polygons[int(region_final)])
# print(all_box_coord[j_cont])
# print(cx_t)
# print(cy_t)
# print(cx_d[i_drop])
# print(cy_d[i_drop])
y_lines = all_box_coord[int(region_final)][0] + np.array(cy_t)
y_lines[y_lines < y_min_d[i_drop]] = 0
# print(y_lines)
arg_min = np.argmin(np.abs(y_lines - y_min_d[i_drop]))
# print(arg_min)
cnt_nearest = np.copy(all_found_texline_polygons[int(region_final)][arg_min])
cnt_nearest[:, 0] = all_found_texline_polygons[int(region_final)][arg_min][:, 0] + all_box_coord[int(region_final)][2]
cnt_nearest[:, 1] = all_found_texline_polygons[int(region_final)][arg_min][:, 1] + all_box_coord[int(region_final)][0]
img_textlines = np.zeros((text_regions_p.shape[0], text_regions_p.shape[1], 3))
img_textlines = cv2.fillPoly(img_textlines, pts=[cnt_nearest], color=(255, 255, 255))
img_textlines = cv2.fillPoly(img_textlines, pts=[polygons_of_drop_capitals[i_drop]], color=(255, 255, 255))
img_textlines = img_textlines.astype(np.uint8)
imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imgray, 0, 255, 0)
contours_combined, hierachy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# print(len(contours_combined),'len textlines mixed')
areas_cnt_text = np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
contours_biggest = contours_combined[np.argmax(areas_cnt_text)]
# print(np.shape(contours_biggest))
# print(contours_biggest[:])
contours_biggest[:, 0, 0] = contours_biggest[:, 0, 0] - all_box_coord[int(region_final)][2]
contours_biggest[:, 0, 1] = contours_biggest[:, 0, 1] - all_box_coord[int(region_final)][0]
contours_biggest = contours_biggest.reshape(np.shape(contours_biggest)[0], np.shape(contours_biggest)[2])
all_found_texline_polygons[int(region_final)][arg_min] = contours_biggest
# all_found_texline_polygons[int(region_final)][arg_min]=contours_biggest
except:
pass
else:
pass
#####for i_drop in range(len(polygons_of_drop_capitals)):
#####for j_cont in range(len(contours_only_text_parent)):
#####img_con=np.zeros((text_regions_p.shape[0],text_regions_p.shape[1],3))
#####img_con=cv2.fillPoly(img_con,pts=[polygons_of_drop_capitals[i_drop] ],color=(255,255,255))
#####img_con=cv2.fillPoly(img_con,pts=[contours_only_text_parent[j_cont]],color=(255,255,255))
#####img_con=img_con.astype(np.uint8)
######imgray = cv2.cvtColor(img_con, cv2.COLOR_BGR2GRAY)
######ret, thresh = cv2.threshold(imgray, 0, 255, 0)
######contours_new,hierachy=cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
#####contours_new,hir_new=return_contours_of_image(img_con)
#####contours_new_parent=return_parent_contours( contours_new,hir_new)
######plt.imshow(img_con)
######plt.show()
#####try:
#####if len(contours_new_parent)==1:
######print(all_found_texline_polygons[j_cont][0])
#####cx_t,cy_t ,_, _, _ ,_,_= find_new_features_of_contoures(all_found_texline_polygons[j_cont])
######print(all_box_coord[j_cont])
######print(cx_t)
######print(cy_t)
######print(cx_d[i_drop])
######print(cy_d[i_drop])
#####y_lines=all_box_coord[j_cont][0]+np.array(cy_t)
######print(y_lines)
#####arg_min=np.argmin(np.abs(y_lines-y_min_d[i_drop]) )
######print(arg_min)
#####cnt_nearest=np.copy(all_found_texline_polygons[j_cont][arg_min])
#####cnt_nearest[:,0]=all_found_texline_polygons[j_cont][arg_min][:,0]+all_box_coord[j_cont][2]
#####cnt_nearest[:,1]=all_found_texline_polygons[j_cont][arg_min][:,1]+all_box_coord[j_cont][0]
#####img_textlines=np.zeros((text_regions_p.shape[0],text_regions_p.shape[1],3))
#####img_textlines=cv2.fillPoly(img_textlines,pts=[cnt_nearest],color=(255,255,255))
#####img_textlines=cv2.fillPoly(img_textlines,pts=[polygons_of_drop_capitals[i_drop] ],color=(255,255,255))
#####img_textlines=img_textlines.astype(np.uint8)
#####imgray = cv2.cvtColor(img_textlines, cv2.COLOR_BGR2GRAY)
#####ret, thresh = cv2.threshold(imgray, 0, 255, 0)
#####contours_combined,hierachy=cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
#####areas_cnt_text=np.array([cv2.contourArea(contours_combined[j]) for j in range(len(contours_combined))])
#####contours_biggest=contours_combined[np.argmax(areas_cnt_text)]
######print(np.shape(contours_biggest))
######print(contours_biggest[:])
#####contours_biggest[:,0,0]=contours_biggest[:,0,0]-all_box_coord[j_cont][2]
#####contours_biggest[:,0,1]=contours_biggest[:,0,1]-all_box_coord[j_cont][0]
#####all_found_texline_polygons[j_cont][arg_min]=contours_biggest
######print(contours_biggest)
######plt.imshow(img_textlines[:,:,0])
######plt.show()
#####else:
#####pass
#####except:
#####pass
return all_found_texline_polygons
def filter_small_drop_capitals_from_no_patch_layout(layout_no_patch, layout1):
drop_only = (layout_no_patch[:, :, 0] == 4) * 1
contours_drop, hir_on_drop = return_contours_of_image(drop_only)
contours_drop_parent = return_parent_contours(contours_drop, hir_on_drop)
areas_cnt_text = np.array([cv2.contourArea(contours_drop_parent[j]) for j in range(len(contours_drop_parent))])
areas_cnt_text = areas_cnt_text / float(drop_only.shape[0] * drop_only.shape[1])
contours_drop_parent = [contours_drop_parent[jz] for jz in range(len(contours_drop_parent)) if areas_cnt_text[jz] > 0.001]
areas_cnt_text = [areas_cnt_text[jz] for jz in range(len(areas_cnt_text)) if areas_cnt_text[jz] > 0.001]
contours_drop_parent_final = []
for jj in range(len(contours_drop_parent)):
x, y, w, h = cv2.boundingRect(contours_drop_parent[jj])
# boxes.append([int(x), int(y), int(w), int(h)])
iou_of_box_and_contoure = float(drop_only.shape[0] * drop_only.shape[1]) * areas_cnt_text[jj] / float(w * h) * 100
height_to_weight_ratio = h / float(w)
weigh_to_height_ratio = w / float(h)
if iou_of_box_and_contoure > 60 and weigh_to_height_ratio < 1.2 and height_to_weight_ratio < 2:
map_of_drop_contour_bb = np.zeros((layout1.shape[0], layout1.shape[1]))
map_of_drop_contour_bb[y : y + h, x : x + w] = layout1[y : y + h, x : x + w]
if (((map_of_drop_contour_bb == 1) * 1).sum() / float(((map_of_drop_contour_bb == 5) * 1).sum()) * 100) >= 15:
contours_drop_parent_final.append(contours_drop_parent[jj])
layout_no_patch[:, :, 0][layout_no_patch[:, :, 0] == 4] = 0
layout_no_patch = cv2.fillPoly(layout_no_patch, pts=contours_drop_parent_final, color=(4, 4, 4))
return layout_no_patch

3
sbb_newspapers_org_image/utils/is_nan.py Normal file
View file

@ -0,0 +1,3 @@
def isNaN(num):
return num != num

252
sbb_newspapers_org_image/utils/marginals.py Normal file
View file

@ -0,0 +1,252 @@
import numpy as np
import cv2
from scipy.signal import find_peaks
from scipy.ndimage import gaussian_filter1d
from .contour import find_new_features_of_contoures, return_contours_of_interested_region
from .resize import resize_image
from .rotate import rotate_image
def get_marginals(text_with_lines, text_regions, num_col, slope_deskew, kernel=None):
mask_marginals=np.zeros((text_with_lines.shape[0],text_with_lines.shape[1]))
mask_marginals=mask_marginals.astype(np.uint8)
text_with_lines=text_with_lines.astype(np.uint8)
##text_with_lines=cv2.erode(text_with_lines,self.kernel,iterations=3)
text_with_lines_eroded=cv2.erode(text_with_lines,kernel,iterations=5)
if text_with_lines.shape[0]<=1500:
pass
elif text_with_lines.shape[0]>1500 and text_with_lines.shape[0]<=1800:
text_with_lines=resize_image(text_with_lines,int(text_with_lines.shape[0]*1.5),text_with_lines.shape[1])
text_with_lines=cv2.erode(text_with_lines,kernel,iterations=5)
text_with_lines=resize_image(text_with_lines,text_with_lines_eroded.shape[0],text_with_lines_eroded.shape[1])
else:
text_with_lines=resize_image(text_with_lines,int(text_with_lines.shape[0]*1.8),text_with_lines.shape[1])
text_with_lines=cv2.erode(text_with_lines,kernel,iterations=7)
text_with_lines=resize_image(text_with_lines,text_with_lines_eroded.shape[0],text_with_lines_eroded.shape[1])
text_with_lines_y=text_with_lines.sum(axis=0)
text_with_lines_y_eroded=text_with_lines_eroded.sum(axis=0)
thickness_along_y_percent=text_with_lines_y_eroded.max()/(float(text_with_lines.shape[0]))*100
#print(thickness_along_y_percent,'thickness_along_y_percent')
if thickness_along_y_percent<30:
min_textline_thickness=8
elif thickness_along_y_percent>=30 and thickness_along_y_percent<50:
min_textline_thickness=20
else:
min_textline_thickness=40
if thickness_along_y_percent>=14:
text_with_lines_y_rev=-1*text_with_lines_y[:]
#print(text_with_lines_y)
#print(text_with_lines_y_rev)
#plt.plot(text_with_lines_y)
#plt.show()
text_with_lines_y_rev=text_with_lines_y_rev-np.min(text_with_lines_y_rev)
#plt.plot(text_with_lines_y_rev)
#plt.show()
sigma_gaus=1
region_sum_0= gaussian_filter1d(text_with_lines_y, sigma_gaus)
region_sum_0_rev=gaussian_filter1d(text_with_lines_y_rev, sigma_gaus)
#plt.plot(region_sum_0_rev)
#plt.show()
region_sum_0_updown=region_sum_0[len(region_sum_0)::-1]
first_nonzero=(next((i for i, x in enumerate(region_sum_0) if x), None))
last_nonzero=(next((i for i, x in enumerate(region_sum_0_updown) if x), None))
last_nonzero=len(region_sum_0)-last_nonzero
##img_sum_0_smooth_rev=-region_sum_0
mid_point=(last_nonzero+first_nonzero)/2.
one_third_right=(last_nonzero-mid_point)/3.0
one_third_left=(mid_point-first_nonzero)/3.0
#img_sum_0_smooth_rev=img_sum_0_smooth_rev-np.min(img_sum_0_smooth_rev)
peaks, _ = find_peaks(text_with_lines_y_rev, height=0)
peaks=np.array(peaks)
#print(region_sum_0[peaks])
##plt.plot(region_sum_0)
##plt.plot(peaks,region_sum_0[peaks],'*')
##plt.show()
#print(first_nonzero,last_nonzero,peaks)
peaks=peaks[(peaks>first_nonzero) & ((peaks<last_nonzero))]
#print(first_nonzero,last_nonzero,peaks)
#print(region_sum_0[peaks]<10)
####peaks=peaks[region_sum_0[peaks]<25 ]
#print(region_sum_0[peaks])
peaks=peaks[region_sum_0[peaks]<min_textline_thickness ]
#print(peaks)
#print(first_nonzero,last_nonzero,one_third_right,one_third_left)
if num_col==1:
peaks_right=peaks[peaks>mid_point]
peaks_left=peaks[peaks<mid_point]
if num_col==2:
peaks_right=peaks[peaks>(mid_point+one_third_right)]
peaks_left=peaks[peaks<(mid_point-one_third_left)]
try:
point_right=np.min(peaks_right)
except:
point_right=last_nonzero
try:
point_left=np.max(peaks_left)
except:
point_left=first_nonzero
#print(point_left,point_right)
#print(text_regions.shape)
if point_right>=mask_marginals.shape[1]:
point_right=mask_marginals.shape[1]-1
try:
mask_marginals[:,point_left:point_right]=1
except:
mask_marginals[:,:]=1
#print(mask_marginals.shape,point_left,point_right,'nadosh')
mask_marginals_rotated=rotate_image(mask_marginals,-slope_deskew)
#print(mask_marginals_rotated.shape,'nadosh')
mask_marginals_rotated_sum=mask_marginals_rotated.sum(axis=0)
mask_marginals_rotated_sum[mask_marginals_rotated_sum!=0]=1
index_x=np.array(range(len(mask_marginals_rotated_sum)))+1
index_x_interest=index_x[mask_marginals_rotated_sum==1]
min_point_of_left_marginal=np.min(index_x_interest)-16
max_point_of_right_marginal=np.max(index_x_interest)+16
if min_point_of_left_marginal<0:
min_point_of_left_marginal=0
if max_point_of_right_marginal>=text_regions.shape[1]:
max_point_of_right_marginal=text_regions.shape[1]-1
#print(np.min(index_x_interest) ,np.max(index_x_interest),'minmaxnew')
#print(mask_marginals_rotated.shape,text_regions.shape,'mask_marginals_rotated')
#plt.imshow(mask_marginals)
#plt.show()
#plt.imshow(mask_marginals_rotated)
#plt.show()
text_regions[(mask_marginals_rotated[:,:]!=1) & (text_regions[:,:]==1)]=4
#plt.imshow(text_regions)
#plt.show()
pixel_img=4
min_area_text=0.00001
polygons_of_marginals=return_contours_of_interested_region(text_regions,pixel_img,min_area_text)
cx_text_only,cy_text_only ,x_min_text_only,x_max_text_only, y_min_text_only ,y_max_text_only,y_cor_x_min_main=find_new_features_of_contoures(polygons_of_marginals)
text_regions[(text_regions[:,:]==4)]=1
marginlas_should_be_main_text=[]
x_min_marginals_left=[]
x_min_marginals_right=[]
for i in range(len(cx_text_only)):
x_width_mar=abs(x_min_text_only[i]-x_max_text_only[i])
y_height_mar=abs(y_min_text_only[i]-y_max_text_only[i])
#print(x_width_mar,y_height_mar,y_height_mar/x_width_mar,'y_height_mar')
if x_width_mar>16 and y_height_mar/x_width_mar<18:
marginlas_should_be_main_text.append(polygons_of_marginals[i])
if x_min_text_only[i]<(mid_point-one_third_left):
x_min_marginals_left_new=x_min_text_only[i]
if len(x_min_marginals_left)==0:
x_min_marginals_left.append(x_min_marginals_left_new)
else:
x_min_marginals_left[0]=min(x_min_marginals_left[0],x_min_marginals_left_new)
else:
x_min_marginals_right_new=x_min_text_only[i]
if len(x_min_marginals_right)==0:
x_min_marginals_right.append(x_min_marginals_right_new)
else:
x_min_marginals_right[0]=min(x_min_marginals_right[0],x_min_marginals_right_new)
if len(x_min_marginals_left)==0:
x_min_marginals_left=[0]
if len(x_min_marginals_right)==0:
x_min_marginals_right=[text_regions.shape[1]-1]
#print(x_min_marginals_left[0],x_min_marginals_right[0],'margo')
#print(marginlas_should_be_main_text,'marginlas_should_be_main_text')
text_regions=cv2.fillPoly(text_regions, pts =marginlas_should_be_main_text, color=(4,4))
#print(np.unique(text_regions))
#text_regions[:,:int(x_min_marginals_left[0])][text_regions[:,:int(x_min_marginals_left[0])]==1]=0
#text_regions[:,int(x_min_marginals_right[0]):][text_regions[:,int(x_min_marginals_right[0]):]==1]=0
text_regions[:,:int(min_point_of_left_marginal)][text_regions[:,:int(min_point_of_left_marginal)]==1]=0
text_regions[:,int(max_point_of_right_marginal):][text_regions[:,int(max_point_of_right_marginal):]==1]=0
###text_regions[:,0:point_left][text_regions[:,0:point_left]==1]=4
###text_regions[:,point_right:][ text_regions[:,point_right:]==1]=4
#plt.plot(region_sum_0)
#plt.plot(peaks,region_sum_0[peaks],'*')
#plt.show()
#plt.imshow(text_regions)
#plt.show()
#sys.exit()
else:
pass
return text_regions

4
sbb_newspapers_org_image/utils/resize.py Normal file
View file

@ -0,0 +1,4 @@
import cv2
def resize_image(img_in, input_height, input_width):
return cv2.resize(img_in, (input_width, input_height), interpolation=cv2.INTER_NEAREST)

85
sbb_newspapers_org_image/utils/rotate.py Normal file
View file

@ -0,0 +1,85 @@
import math
import imutils
import cv2
def rotatedRectWithMaxArea(w, h, angle):
if w <= 0 or h <= 0:
return 0, 0
width_is_longer = w >= h
side_long, side_short = (w, h) if width_is_longer else (h, w)
# since the solutions for angle, -angle and 180-angle are all the same,
# if suffices to look at the first quadrant and the absolute values of sin,cos:
sin_a, cos_a = abs(math.sin(angle)), abs(math.cos(angle))
if side_short <= 2.0 * sin_a * cos_a * side_long or abs(sin_a - cos_a) < 1e-10:
# half constrained case: two crop corners touch the longer side,
# the other two corners are on the mid-line parallel to the longer line
x = 0.5 * side_short
wr, hr = (x / sin_a, x / cos_a) if width_is_longer else (x / cos_a, x / sin_a)
else:
# fully constrained case: crop touches all 4 sides
cos_2a = cos_a * cos_a - sin_a * sin_a
wr, hr = (w * cos_a - h * sin_a) / cos_2a, (h * cos_a - w * sin_a) / cos_2a
return wr, hr
def rotate_max_area_new(image, rotated, angle):
wr, hr = rotatedRectWithMaxArea(image.shape[1], image.shape[0], math.radians(angle))
h, w, _ = rotated.shape
y1 = h // 2 - int(hr / 2)
y2 = y1 + int(hr)
x1 = w // 2 - int(wr / 2)
x2 = x1 + int(wr)
return rotated[y1:y2, x1:x2]
def rotation_image_new(img, thetha):
rotated = imutils.rotate(img, thetha)
return rotate_max_area_new(img, rotated, thetha)
def rotate_image(img_patch, slope):
(h, w) = img_patch.shape[:2]
center = (w // 2, h // 2)
M = cv2.getRotationMatrix2D(center, slope, 1.0)
return cv2.warpAffine(img_patch, M, (w, h), flags=cv2.INTER_CUBIC, borderMode=cv2.BORDER_REPLICATE)
def rotyate_image_different( img, slope):
# img = cv2.imread('images/input.jpg')
num_rows, num_cols = img.shape[:2]
rotation_matrix = cv2.getRotationMatrix2D((num_cols / 2, num_rows / 2), slope, 1)
img_rotation = cv2.warpAffine(img, rotation_matrix, (num_cols, num_rows))
return img_rotation
def rotate_max_area(image, rotated, rotated_textline, rotated_layout, angle):
wr, hr = rotatedRectWithMaxArea(image.shape[1], image.shape[0], math.radians(angle))
h, w, _ = rotated.shape
y1 = h // 2 - int(hr / 2)
y2 = y1 + int(hr)
x1 = w // 2 - int(wr / 2)
x2 = x1 + int(wr)
return rotated[y1:y2, x1:x2], rotated_textline[y1:y2, x1:x2], rotated_layout[y1:y2, x1:x2]
def rotation_not_90_func(img, textline, text_regions_p_1, thetha):
rotated = imutils.rotate(img, thetha)
rotated_textline = imutils.rotate(textline, thetha)
rotated_layout = imutils.rotate(text_regions_p_1, thetha)
return rotate_max_area(img, rotated, rotated_textline, rotated_layout, thetha)
def rotation_not_90_func_full_layout(img, textline, text_regions_p_1, text_regions_p_fully, thetha):
rotated = imutils.rotate(img, thetha)
rotated_textline = imutils.rotate(textline, thetha)
rotated_layout = imutils.rotate(text_regions_p_1, thetha)
rotated_layout_full = imutils.rotate(text_regions_p_fully, thetha)
return rotate_max_area_full_layout(img, rotated, rotated_textline, rotated_layout, rotated_layout_full, thetha)
def rotate_max_area_full_layout(image, rotated, rotated_textline, rotated_layout, rotated_layout_full, angle):
wr, hr = rotatedRectWithMaxArea(image.shape[1], image.shape[0], math.radians(angle))
h, w, _ = rotated.shape
y1 = h // 2 - int(hr / 2)
y2 = y1 + int(hr)
x1 = w // 2 - int(wr / 2)
x2 = x1 + int(wr)
return rotated[y1:y2, x1:x2], rotated_textline[y1:y2, x1:x2], rotated_layout[y1:y2, x1:x2], rotated_layout_full[y1:y2, x1:x2]

1752
sbb_newspapers_org_image/utils/separate_lines.py Normal file
View file

File diff suppressed because it is too large Load diff

2
setup.py
View file

@ -16,7 +16,7 @@ setup(
install_requires=install_requires,
entry_points={
'console_scripts': [
'eynollah=sbb_newspapers_org_image.eynollah:main',
'eynollah=sbb_newspapers_org_image.cli:main',
# 'ocrd-eynollah=eynollah.ocrd_cli:cli',
]
},

BIN
tests/resources/kant_aufklaerung_1784_0020.tif Normal file
View file

Binary file not shown.

7
tests/test_smoke.py Normal file
View file

@ -0,0 +1,7 @@
def test_utils_import():
import sbb_newspapers_org_image.utils
import sbb_newspapers_org_image.utils.contour
import sbb_newspapers_org_image.utils.drop_capitals
import sbb_newspapers_org_image.utils.drop_capitals
import sbb_newspapers_org_image.utils.is_nan
import sbb_newspapers_org_image.utils.rotate