Separating characters manually
Eli Bluehatseo.com is spending a week breaking captchas after they ddos’d his server. I’m interested.
I’m going to explain how I would separate the characters in an image after removing the noise and lines and stuff. It’s funny this post is going to be pretty damn short 
but I need time to work out the post with neural networks. This will just be the theory of separating characters and I’ll post code next time.
So first things first we need to scan through the vertical lines until we hit a pixel. From that pixel we flood fill. Yep. That’s it. As long as it’s a custom flood fill routine it will give us a start and end point from which we can fit a polygon or rectangle around the letter and extract it out. Now just de-rotate it and throw it at the neural network I don’t have yet.
Tuesday, March 18th, 2008
