OpenCV：如何从黑白图像的特定部分删除文本

Question

Good morning.

I am developing an application in Python to detect the frames of a manga page, and that detects the speech bubbles to erase the text. Example: 1 : 1st Image:

Ideal result 1:

However, OCV has difficulties to detect the text formed in Japanese or Korean, so I thought that perhaps it was easier to find an algorithm that recognized the shape of the speech bubbles and filled their interior with white, eliminating the text.

I would also like to divide the manga pages by their frames, although this is much simpler, because they are usually square and have a frame that makes them easily detectable.

The problem is that I dont have much idea about OCV or Tesseract , and I wouldnt even know where and how to start.

This user repository is what I am using but doesnt work too well for what I want...

https://github.com/DivyaKrishnani/Working-with-Text-on-Images

Answer 1

I can think of the following methods:

1. You can run an OCR to detect Korean texts and their corresponding box. Then you can paint the whole bounding box with white colour to remove these Korean words. There are free OCR python libraries available like Pytesseract, Keras-OCR, easy OCR, EAST text detection. Paid ones by Microsoft Azure, Amazon and Google.

2. You can treat each character as an object and train an object detection module (EfficientDet or YOLO) to detect the same. You will get the bounding box and do the same steps as mentioned in previous point.