How to improve results of tesseract to recognize numbers

I am working on iPhone application that recognizes ISBN numbers (ISBN: 978-83-7380-900-0) I use tesseract for this, but it is not working very well. I can see other applications, using same engine to work better.

to limit the characters i use this config line: tess->SetVariable("tessedit_char_whitelist", "SN:0123456789X-"); so all "I" are converted to "1", and "B" to 8. Using this it wont make mistake with those letters, whick are not important to me. After that i use regular expression to find the correct part of recognized text.

I also crop the image, so tesseract recognizes only part of the image, where isbn is visible (i placed color rect on camera overlay, so user have to place code in correct place) I also resize开发者_如何学Go the image to 1000px width (also tried other sizes)

It works quite well when the light is excellent, but it is really hard to recognize correctly when the lighting isn't perfect.

The last digit of isbn number is a control sum.

What can I do to make it work better? Is there any way to say tesserect to recognize text only in given regular expression? Maybe i should do something with image first?

Sample images, that are not recognized correctly:

http://img412.imageshack.us/i/img0367si.jpg/

http://img264.imageshack.us/i/img0361d.jpg/

---

i used some pre processing and results are quite good at the moment. (normalisation and auto threshold) i do it with simple image processing for iphone. I am also making images with 2x zoom. The only problem is that it is really easy to make fuzzy image, and then it is imossible to recognize anything. The process of ocr recognition takes aproximately 4s. and it works on image 300 x 109 : 19KB I am just wondering what can i do do make it working faster, so i would be able to take several images and process all of them.