def get_table_text_dict(self, test_gerber=False, save_dict=False, highlight_readable_paras=None,
v_cut_save_path=None):
"""
save table text dict in a list
"""
print('I am working on extracting table information dictionary')
table_num = 0
table_loc = []
table_dicts_group = []
pure_text_dict_group = []
# adjust contrast
contrast_img = cv2.addWeighted(self.image, 1.3, self.blank_image, 1 - 1.3, 5)
highlight_step_1 = self.blank_image.copy()
gerber_file = get_file_type(self.image_path, test_gerber)
if gerber_file:
imgs, table_coo = read_table.extract_table_from_img(self.image_path)
pure_table_image = np.zeros([self.image_height + self.soft_margin, self.image_width + self.soft_margin, 3],
self.image.dtype)
for table in imgs:
if not intersection_lines_detection(table):
table_locate_info = self.get_pure_table_region(table_coo, table_num)
# filling small table region prepare big table contours detector
cv2.rectangle(pure_table_image, (table_locate_info[0], table_locate_info[1]),
(table_locate_info[0] + table_locate_info[2],
table_locate_info[1] + table_locate_info[3]),
(0, 0, 255), thickness=-1)
table_loc.append(table_locate_info)
# cv2.rectangle(self.covered_text_image, (table_locate_info[0], table_locate_info[1]),
# (table_locate_info[0] + table_locate_info[2],
# table_locate_info[1] + table_locate_info[3]),
# (0, 0, 0), thickness=-1)
table_num += 1
else:
table_num += 1
continue
# print(self.covered_text_image.shape)
# f = open('4S7MD161A0_table_loc.pkl', 'wb')
# pickle.dump(table_loc, f)
# f.close()
# cv2.imwrite('pure_pure_table.png', pure_table_image)
dilate_kernel = cv2.getStructuringElement(cv2.MORPH_OPEN, (7, 7))
dilate_image = cv2.dilate(pure_table_image, dilate_kernel, iterations=2)
binary_table_region = get_binary(dilate_image, my_threshold=[45, 255])
table_edge_condition, table_region_contours = find_text_region(binary_table_region, cv2.RETR_EXTERNAL)
print('I am working on pure text OCR')
o_img = self.image.copy()
background_color = get_dominant_color(o_img)
for edge_num in range(len(table_edge_condition)):
# draw big table contours
cv2.drawContours(o_img, table_edge_condition, edge_num, background_color, thickness=3)
pure_text_dict_group = pure_text_region(o_img, background_color, self.blank_image)
# cv2.imwrite('pure_table.png', binary_table_region)
print('I am working on tables OCR')
i = 0
for edge_condition in table_edge_condition:
sorted_edge_condition = sorted(edge_condition.tolist())
min_point = sorted_edge_condition[0]
max_point = sorted_edge_condition[-1]
cut_table = self.image[min_point[1]:max_point[1], min_point[0]:max_point[0]]
c_x, c_y, c_z = cut_table.shape
if c_x and c_y and c_z:
table_ram = 'table_RAM' + str(i) + '.png'
# table_ram = 'table_RAM.png'
# print(table_ram)
cv2.imwrite(table_ram, cut_table)
i += 1
table_ram = ocr_preprocessed(table_ram)
iso_table_dict = self.ocr_detector(table_ram, self.blank_image, self.soft_margin, min_point)
if iso_table_dict:
table_dicts_group.append(iso_table_dict)
# os.remove(table_ram)
if not table_dicts_group:
print('not text')
v_cut_detector(self.image, v_cut_save_path)
elif not gerber_file:
print('I am not gerber file using ocr')
self.soft_margin = 0
try:
iso_table_dict = self.ocr_detector(self.image_path, self.blank_image, self.soft_margin, [0, 0])
except Exception as e:
print('OCR sever failed: {} Use backup solution'.format(e))
iso_table_dict = self.ocr_detector(self.image_path, self.blank_image, self.soft_margin, [0, 0],
ocr_type='you_dao')
if iso_table_dict:
table_dicts_group.append(iso_table_dict)
table_dicts_group.extend(pure_text_dict_group)
if save_dict:
# save table dictionary in pkl file
print('save dictionary into pickle file')
f_name = get_file_name(self.image_path)
f_extension = get_extension(self.image_path)
file = open(f_name + f_extension + '.pkl', 'wb')
print('save path is:{}'.format(f_name + f_extension + '.pkl'))
table_save_as_list = text_dict2text_list(table_dicts_group)
print(table_save_as_list)
pickle.dump(table_save_as_list, file)
file.close()
if highlight_readable_paras:
print('highlight step 1')
highlight_step_1 = cv2.addWeighted(contrast_img, 0.8, self.blank_image, 0.2, 3)
f_name = get_file_name(self.image_path)
f_extension = get_extension(self.image_path)
cv2.imwrite(f_name + '_Marked_' + f_extension, highlight_step_1)
return table_dicts_group, highlight_step_1
def get_table_text_dict(self, highlight_readable_paras=True):
"""
save table text dict in a list
"""
print('I am working on extracting GM2 information dictionary')
def _draw_rectangle(_image, color):
cv2.rectangle(_image, (table_locate_info[0], table_locate_info[1]),
(table_locate_info[0] + table_locate_info[2],
table_locate_info[1] + table_locate_info[3]),
color,
thickness=-1)
table_num = 0
table_dicts_group = []
image_save_path = ''
contrast_img = cv2.addWeighted(self.image, 1.3, self.blank_image,
1 - 1.3, 5)
imgs, table_coo = read_table.extract_table_from_img(self.image_path)
pure_table_image = np.zeros([
self.image_height + self.soft_margin,
self.image_width + self.soft_margin, 3
], self.image.dtype)
os.remove(self.image_path)
for table in imgs:
if not intersection_lines_detection(table):
table_locate_info = self.get_pure_table_region(
table_coo, table_num)
# filling small table region prepare big table contours detector
_draw_rectangle(pure_table_image, color=(0, 0, 255))
_draw_rectangle(self.covered_text_image, color=(0, 0, 0))
table_num += 1
else:
table_num += 1
continue
dilate_kernel = cv2.getStructuringElement(cv2.MORPH_OPEN, (7, 7))
dilate_image = cv2.dilate(pure_table_image,
dilate_kernel,
iterations=2)
binary_table_region = get_binary(dilate_image, my_threshold=[45, 255])
cv2.imwrite('binary_table_region.png', binary_table_region)
table_edge_condition, table_region_contours = find_text_region(
binary_table_region, cv2.RETR_EXTERNAL)
print('I am working on tables OCR')
covered_text_image = cv2.subtract(self.covered_text_image, self.image)
for edge_condition in table_edge_condition:
sorted_edge_condition = sorted(edge_condition.tolist())
min_point = sorted_edge_condition[0]
max_point = sorted_edge_condition[-1]
cut_table = self.image[min_point[1]:max_point[1],
min_point[0]:max_point[0]]
covered_text_table = covered_text_image[min_point[1]:max_point[1],
min_point[0]:max_point[0]]
c_x, c_y, c_z = cut_table.shape
if c_x and c_y and c_z:
table_ram = 'table_RAM.png'
cv2.imwrite(table_ram, cut_table)
min_max_points_group = find_min_max_points_group(
covered_text_table, min_point)
iso_table_dict = self.ocr_detector(table_ram, self.blank_image,
self.soft_margin, min_point)
if iso_table_dict:
grouped_small_tables = group_gerber_ocr_text(
min_max_points_group, iso_table_dict,
self.gerber_boxes)
table_dicts_group.append(
list(grouped_small_tables.values()))
os.remove(table_ram)
if highlight_readable_paras:
print('highlight step 1')
highlight_step_1 = cv2.addWeighted(contrast_img, 0.8,
self.blank_image, 0.2, 3)
f_name = get_file_name(self.gm_path)
file_name = f_name.split('/')[-1]
f_extension = get_extension(self.image_path)
tmp_file_path = '/data/fastprint/tmp-files'
current_time = datetime.now().strftime('%Y%m%d%H%M%S')
save_path = "{tmp_file_path}/{current_time}".format(
tmp_file_path=tmp_file_path, current_time=current_time)
# 如果文件夹不存在，创建文件夹
if not os.path.exists(save_path):
os.mkdir(save_path)
# 图片保存的路径
image_save_path = "{save_path}/{file_name}_Marked_{f_extension}".format(
save_path=save_path,
file_name=file_name,
f_extension=f_extension)
cv2.imwrite(image_save_path, highlight_step_1)
return table_dicts_group, image_save_path
def get_table_text_dict(self,
save_dict_path=None,
save_dict=False,
highlight_readable_paras=None):
"""
save table text dict in a list
"""
print('I am working on extracting table information dictionary')
table_num = 0
table_dict = {}
table_keys_list = []
table_dicts_group = []
# adjust contrast
contrast_img = cv2.addWeighted(self.image, 1.3, self.blank_image,
1 - 1.3, 5)
imgs, table_coo = read_table.extract_table_from_img(self.image_path)
highlight_step_1 = self.blank_image.copy()
pure_table_image = np.zeros([
self.image_height + self.soft_margin,
self.image_width + self.soft_margin, 3
], self.image.dtype)
print('I am working on OCR')
for table in imgs:
# print('working on region {}'.format(table_num))
if not intersection_lines_detection(table):
table_key, table_value = self.no_intersection_tables_ocr(
table, table_coo, table_num, highlight_readable_paras)
cv2.rectangle(
pure_table_image, (table_key[0], table_key[1]),
(table_key[0] + table_key[2], table_key[1] + table_key[3]),
(0, 0, 255),
thickness=4)
table_dict[tuple(table_key)] = table_value
table_keys_list.append(table_key)
table_num += 1
else:
table_num += 1
continue
# image_copy = self.image.copy()
binary_table_region = get_binary(pure_table_image)
table_edge_condition, table_region_contours = find_text_region(
binary_table_region, cv2.RETR_EXTERNAL)
# cv2.drawContours(image_copy, table_region_contours, -1, (0, 255, 0), 12)
# cv2.imwrite('pure_table_region.png', image_copy)
# cv2.imshow('pure table region', pure_table_image)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# group the discrete small tables in a table
# table_edge_condition = get_iso_table_condition(table_keys_list)
# print('return table text information dictionary')
for edge_condition in table_edge_condition:
represent_min_point = sorted(edge_condition.tolist())[0]
represent_max_point = sorted(edge_condition.tolist())[-1]
min_edge_x, min_edge_y = represent_min_point
max_edge_x, max_edge_y = represent_max_point
iso_table_dict = {}
for key in table_keys_list:
if min_edge_x <= key[0] < max_edge_x and min_edge_y <= key[
1] < max_edge_y:
iso_table_dict[tuple(key)] = table_dict[tuple(key)]
table_dicts_group.append(iso_table_dict)
if save_dict:
# save table dictionary in pkl file
file = open(save_dict_path, 'wb')
pickle.dump(table_dict, file)
file.close()
if highlight_readable_paras:
print('highlight step 1')
highlight_step_1 = cv2.addWeighted(contrast_img, 0.8,
self.blank_image, 0.2, 3)
# f_name = get_file_name(save_highlight_path)
# f_extension = get_extension(save_highlight_path)
# cv2.imwrite('test_highlight_step1_marked.png', highlight_step_1)
return table_dicts_group, highlight_step_1