def mask_tables(page_imge, check=False):
try:
table_image = cv2.imread(page_imge, 0)
page_img = cv2.imread(page_imge)
table_image = clean_image(table_image)
page_img = clean_image(page_img)
image_width, image_height = table_image.shape[1], table_image.shape[0]
if check:
cv2.imwrite('bg_org.png', table_image)
except Exception as e:
log_error(
"Service TableExtractor Error in loading background html image",
app_context.application_context, e)
return None, None
# To do recognize and mask images berfor finding tables
#table_image = mask_image(table_image, img_df, image_width, image_height, app_context.application_context,
# margin=0, fill=255)
#check = True
if check:
cv2.imwrite('bg_org_masked.png', table_image)
try:
tables = TableRepositories(table_image).response['response']['tables']
except Exception as e:
log_error("Service TableExtractor Error in finding tables",
app_context.application_context, e)
return None, None
try:
rects = RectRepositories(table_image)
lines, _ = rects.get_tables_and_lines()
except Exception as e:
log_error("Service TableExtractor Error in finding lines",
app_context.application_context, e)
return None, None
#line_regions = get_regions(lines, 'LINE')
tables_regions = get_regions(tables, 'TABLE')
masked_image = mask_image(page_img,
tables,
image_width,
image_height,
app_context.application_context,
margin=0,
fill=255)
### add line regions
#return masked_image, line_regions + tables_regions
return masked_image, tables_regions
def detect_text_per_file(image_paths,
network,
text_threshold,
low_text_threshold,
link_threshold,
img_class="single_col"):
in_dfs = []
number_of_pages = len(image_paths)
if img_class == "double_col":
number_of_pages = 1
image_paths = [image_paths]
t = time.time()
for image_path in image_paths:
image = cv2.imread(image_path)
image = clean_image(image)
# if img_class == "double_col":
# image = image_path
# else:
# image = imgproc.loadImage(image_path)
if network:
bboxes, polys, score_text = test_net(image, text_threshold,
link_threshold,
low_text_threshold, args.cuda,
args.poly, refine_net)
else:
bboxes, polys, score_text = test_net(image, text_threshold,
link_threshold,
low_text_threshold, args.cuda,
args.poly, None)
column_names = ["x1", "y1", "x4", "y4", "x2", "y2", "x3", "y3"]
df = pd.DataFrame(columns=column_names)
for index, box in enumerate(bboxes):
#for index, box in enumerate(polys):
poly = np.array(box).astype(np.int32).reshape((-1))
df.at[index, 'x1'] = int(poly[0])
df.at[index, 'y1'] = int(poly[1])
df.at[index, 'x2'] = int(poly[2])
df.at[index, 'y2'] = int(poly[3])
df.at[index, 'x3'] = int(poly[4])
df.at[index, 'y3'] = int(poly[5])
df.at[index, 'x4'] = int(poly[6])
df.at[index, 'y4'] = int(poly[7])
color = (255, 0, 0)
thickness = 2
cv2.rectangle(image, (int(poly[0]), int(poly[1])),
(int(poly[4]), int(poly[5])), color, thickness)
#cv2.imwrite("/home/naresh/word_compare3/"+str(uuid.uuid4())+".jpg", image)
#in_df = convert_to_in_df(df)
in_dfs.append(df)
time_taken = time.time() - t
time_take_per_page = time_taken / number_of_pages
message = 'Time taken for text detection is ' + str(
time_taken) + '/' + str(number_of_pages) + 'time per page : ' + str(
time_take_per_page)
log_info(message, app_context.application_context)
return in_dfs