0

I'm trying to design a CAPTCHA recognition algorithm using Python, OpenCV and Tesseract. The problem is the CAPTCHA digits are misaligned and randomly clustered within the image. Pytesseract fails and returns empty lists most of the time. The data looks like: this should return 41332 this should return 35545

The relatively flatter CAPTCHA's where the digits are in a line get detected better. How do I solve this? How to detect, crop and realign the digits in images like these to make it easier for tesseract(if needed to be used) to detect them?

Siddhesh Acharekar
  • 37
  • 1
  • 1
  • 4
  • 2
    Well captcha's are meant **not** to be recognized by computer vision. Otherwise it would defeat the entire purpose of captcha's. – Willem Van Onsem Feb 06 '17 at 10:36
  • I know, this is for a relatively one dimensional problem though. Have a list of images that are 5 digit numbers clearly written in the same font and size ONLY misaligned and clustered randomly. So I'm looking for some help preprocessing each image that would make it easier for Tesseract. Or any other approach involving ML. Anything that helps – Siddhesh Acharekar Feb 06 '17 at 10:43

0 Answers0