If you do a google search "Matlab Hough transform code" you'll get some code from mathworks.com (user contributed). If you want specific feature extraction code, you need to decide what kind of features you would use. In general, if you normalize all images so that there are no margins in the character images (the characters are touching the borders of the character image files, then simply taking a few pixel wide border of these images (bw images) and using these as input vectors to something like RBF classifiers should work pretty well. I've used it and works >80% if the time. That should solve your classification problem. If you need to decide which features to use, you can use SIFT features.

If you want code to "extract" these characters out of images that's a whole new OCR problem.