texts formatting with maintaining texts alignment - python-3.x

Question

I am using open-source tesseract OCR to extract texts from the images. I need texts with proper space preserved and aligned for some information extraction tasks. I have used the tesseract .image_to_data() method. This method outputs the recognized texts along with its bounding box and confidence score.
After performing sorting I get all the texts on a string in the order they exist in the image but not formatted in the way that I want. One of my test images is
My code:
import cv2
import pytesseract
import argparse
import time
from pytesseract import Output
from utils.config import ENG_CONFIG
def check_confidence(df):
df = df[df['conf'] != -1]
return df
class OCR:
def __init__(self, conf, image):
self.conf = conf
self.image = image
def get_ocr_output(self):
result = pytesseract.image_to_data(self.image, config=self.conf, output_type=Output.DATAFRAME)
result = check_confidence(result)
lines, formatted_string = self.parse_texts(result)
return lines, formatted_string
def parse_texts(self, df):
img = self.image.copy()
parsed_info = []
start = time.time()
for row in df.itertuples(): # for index, row in df.iterrows()
loc = (row.left, row.top, row.width, row.height)
# draw box on the image
cv2.rectangle(img, (loc[0], loc[1]), (loc[0] + loc[2], loc[1] + loc[3]), (255, 255, 0), 2)
text = row.text
conf = row.conf
row_info = {'box': loc, 'text': text, 'conf': conf}
parsed_info.append(row_info)
end = time.time()
print("\n[INFO]: Total time to read the dataframe row by row is : {} seconds.".format(end - start))
print("\n[INFO]: Parsed Information before sorting boxes is: \n{}".format(parsed_info))
parsed_info = self.sort_box_vertically(parsed_info)
lines = self.get_lines_vertically(parsed_info)
lines = self.sort_box_horizontally(lines)
print("\n[INFO]: Parsed Information after sorting boxes horizontally-vertically is : \n{}".format(lines))
# formatted_strings = self.format_texts(lines)
formatted_strings = self.format_texts_by_reserving_spaces(lines)
print("\n[INFO] : Text for the image is : \n{}".format(formatted_strings))
cv2.imshow("text box", img)
cv2.waitKey(0)
return lines, formatted_strings
def sort_box_vertically(self, parsed_info):
"""
Method to sort the boxes vertically
:param parsed_info:
:return:
"""
for counter in range(len(parsed_info) - 1):
for num in range((len(parsed_info) - 1) - counter):
if self.get_box_center(parsed_info[num]['box'])[1] > self.get_box_center(parsed_info[num + 1]['box'])[
1]:
temp = parsed_info[num + 1]
parsed_info[num + 1] = parsed_info[num]
parsed_info[num] = temp
return parsed_info
def get_lines_vertically(self, parsed_info):
"""
Method to separate the lines vertically
:param parsed_info:
:return:
"""
lines = []
while len(parsed_info) != 0:
ref_box = parsed_info[0]['box']
line = [parsed_info[0]]
for i in range(1, len(parsed_info)):
if self.is_intersects(ref_box, parsed_info[i]['box']) or self.is_intersects(parsed_info[i]['box'],
ref_box):
line.append(parsed_info[i])
for i in range(len(line)):
parsed_info.remove(line[i])
lines.append(line)
return lines
def is_intersects(self, box1, box2, margin=0):
x_1, y_1, w_1, h_1 = box1
x_2, y_2, w_2, h_2 = box2
# check if box2 intersect with box1 by height
h1, k1 = self.get_box_center(box1)
h2, k2 = self.get_box_center(box2)
if int(k2) in range(int(y_1 + margin * h_1), int((y_1 + h_1 - h_1 * margin))) or int(k1) in range(
int(y_2 + margin * h_2), int((y_2 + h_2 - h_2 * margin))):
return True
return False
def get_box_center(self, box):
"""
Method to get the center of box (h,k)
:param box: box as tuple (x, y, w, h)
:return:
"""
x, y, w, h = box
x1 = x
y1 = y
x2 = x + w
y2 = y + h
# Find the center of the box
h, k = int((x1 + x2) / 2), int((y1 + y2) / 2)
return h, k
def format_texts(self, lines):
"""
Method to return the formatted texts from the image
:param lines:
:return:
"""
formatted_string = ""
for line in lines:
for word in line:
# print(word)
formatted_string += " " + word['text']
formatted_string += '\n'
return formatted_string
def format_texts_by_reserving_spaces(self, lines):
"""
An efficient Method to return the formatted texts from the image
:param lines:
:return:
"""
formatted_string = ""
for line in lines:
flag = 0
for word in line:
# print(word)
word_start = word['box'][0]
word_end = word['box'][0] + word['box'][2]
formatted_string += " " * (word_start - flag - 1) + word['text']
flag = word_end
formatted_string += '\n'
return formatted_string
def sort_box_horizontally(self, lines):
final_lines = []
for line in lines:
for counter in range(len(line) - 1):
for num in range((len(line) - 1) - counter):
# If x1 of first line is greater than second line
if self.get_box_center(line[num]['box'])[0] > self.get_box_center(line[num + 1]['box'])[0]:
temp = line[num + 1]
line[num + 1] = line[num]
line[num] = temp
final_lines.append(line)
return final_lines
def sort_box_horizontally_example(self, line_boxes):
"""
Method to sort the boxes horizontally
:param boxes: list of tuple of boxes eg:[((,),(,))]
:return:
"""
final_lines = []
for line in line_boxes:
for counter in range(len(line) - 1):
for num in range((len(line) - 1) - counter):
# If x1 of first line is greater than second line
if line[num][0][0] > line[num + 1][0][0]:
temp = line[num + 1]
line[num + 1] = line[num]
line[num] = temp
final_lines.append(line)
return final_lines
def sorting_example(self):
"""
Sorting array using bubble sort
:return:
"""
list_ = [3, 6, 8, 44, 77, 2, 44, 556, 66, 565, 34]
for i in range(len(list_) - 1):
for j in range((len(list_) - 1) - i):
if list_[j] > list_[j + 1]:
temp = list_[j]
list_[j] = list_[j + 1]
list_[j + 1] = temp
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('-i', '--image',
default="./intelligent_information_extractor/images/img5.png",
help="Path to input image to be OCR'd")
args = parser.parse_args()
image = cv2.imread(args.image)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
ocr = OCR(ENG_CONFIG, image)
results = ocr.get_ocr_output()
The output it produces can be found here
While formatting texts which ocr outputs, I have converted a pixel to a single character. But it creates the text string where texts are separated by large whitespace.
So, How to decrease the whitespace to properly format the output texts by preserving all the spaces and alignment?