A simple fast ML model trained over a defined captcha dataset that is capable of deciphering captcha images.
The dataset was provided as a part of assignment for the course CS771 at the Indian Institute of Technology, Kanpur, during Summer Term 2022.
The dataset consists of 2000 images with 4 hex characters in each.
Each image has a light background color with comparatively darker color for the characters.
There are some lines in the background which are guarenteed to be not as thick as the characters itself.
The characters are oriented only in -30 degrees, -20 degrees, -10 degrees, 0 degrees, 10 degrees, 20 degrees and 30 degrees.
The original intention was to figure out whether the value in the captcha is an odd or even number and the labels were also provided for such.
The Label.txt file available in the dataset was created by the owner for extending the project to cracking the entire captcha.
Link to the training dataset used - https://drive.google.com/file/d/1AVihDYXcvMrf4aftXJkZHIR8h1v0XFS_/view?usp=sharing