Computer Vision Done Quick
Part of the joy of working on a brand new project is exploring just how much work it will be to solve new problems. One of the requested features is to build in a decent document scanner, to cut down on requiring users to send faxes of documentation.
After looking at some of the open source libraries for the sort of work on iOS, I decided to just implement a pair of CIDetectors to scan for rectangles and text, then see how the results came out. They’re not perfect, and there are some quirks that I’m working out, but the result are decent. With some good UI design to allow the user to adjust for minor skew and hopefully a few tricks to improve the quality of the results, I think that an optimized version of the out of the box solution will be good enough to ship with the first release.
I’m already collecting interesting ideas on just what I could do using OpenCV, and taking inspiration from some of the very cool work in the Dropbox document scanner to make an even better recognizer.