Word Verification Reader, Update #3

I finished revamping the bitmap-to-line-segments algorithm to deal with "branching" ... where a horizontal line has multiple "columns" descending below it. A little bit of state-saving and recursion made it all work magically. The algorithm needs a couple tweaks though; nothing major.

Next, instead of simply depicting the lines and points visually, I'll start storing the coordinates. I can then extend and smooth the lines with sub-pixel accuracy.

The asterisks and hashes might look a little haphazard. Each asterisk represents the end-point to a line segment. The hashes are simply points - where the tolerance for a continuancy was exceeded, but the result was only a single line - in which case the hash is the center point of that horizontal line.

So, if the day comes when this works reliably, what then? In its current form, I could save the Word Verification images to disk, and have it print to the console the associated letters. But, besides giving me the satisfaction of a job well done, that really doesn't accomplish anything. Being written in C, it is probably mostly outside the reach of Joe-Spammer, but publishing the algorithm is all I would need to do to leave my mark on this world. An implementation in Perl and Python would probably help that image quite a bit as well. Hmm. If I play this wrong, I'll be making my own Nobel Prize (the Jones Prize?) as a means of quelling my conscience. While TNT went to use in killing people, my program resulted in the down-fall of Internet communication.