1

I want to detect the percentage of the area the text blocks are using on an image. The idea is to reject images that have more than 40% text on it. I saw a very informative and detailed post in detecting text here. The link is having C++ used. I think I can get the idea to use it in python.

However, am not sure what would be the best way to measure the percentage of the area it is using. Is there any implementation of something similar that we could use? I am just getting started with Cv.

I am getting my text as below using python. This is a code I found to work from a forum online. 

import cv2

def captch_ex(file_name):
    img = cv2.imread(file_name)
    img_final = cv2.imread(file_name)
    img2gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    ret, mask = cv2.threshold(img2gray, 180, 255, cv2.THRESH_BINARY)
    image_final = cv2.bitwise_and(img2gray, img2gray, mask=mask)
    ret, new_img = cv2.threshold(image_final, 180, 255, cv2.THRESH_BINARY) 



    kernel = cv2.getStructuringElement(cv2.MORPH_CROSS, (3,3)) 
    dilated = cv2.dilate(new_img, kernel, iterations=9)  

    contours, hierarchy = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

    for contour in contours:

        [x, y, w, h] = cv2.boundingRect(contour)


        if w < 35 and h < 35:
            continue

        cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 255), 2)


    cv2.imshow('captcha_result', img)
    cv2.waitKey()

file_name = 'my_image.jpg'
captch_ex(file_name)

However, how do I proceed further?

esafwan
  • 17,311
  • 33
  • 107
  • 166
  • Have you read the content of the link? FYI There is another answer where the C++ code is rewritten in Python. What have you tried? – DisappointedByUnaccountableMod Jun 07 '17 at 21:44
  • @barny I have tried text detection a while ago. That part I am confident, but with respect to measuring area am totally blank. I haven't tried anything with respect to measuring area, hence I have no clue where to start. So if get some clue, I want to try it. – esafwan Jun 07 '17 at 22:06
  • @barny Added the code that worked for me, which I found online. – esafwan Jun 07 '17 at 22:26

1 Answers1

1

A good starting point is to

  1. Calculate the total area (w*h) of bounding boxes. This corresponds to the total area of the image covered with text.

  2. Compare that with the total number of pixels in the image (rows*cols)

Do note that this value will not be accurate as it is, because there can be different font sizes and line spacings. To determine the exact margin where 40% of the image is covered with text, you should find the threshold by picking several images with around 40% covered with text, both above and below that amount, as identified by humans.

Totoro
  • 3,398
  • 1
  • 24
  • 39