Bank Notes Automatic Identification System

As we all know, piles of notes and bills are handled and processed everyday in all banks. And the last step is the notes scanned to digital image. But now the information of them still needs to be input manually. So mechanical! The situation can be changed, based on my thesis, those operation all can be done by a computer.Combined with DSP and Pattern Recognise technology, the numerical characters, the bank name and bank no. can be verified and obtained. I adoped the method of improve BP network combined with broad matching to get the information. The system self-learning ability can be enhanced by the improved S function suitable to identify numerical charaters as the nerve cell. And In arithmetic, I adoped dynamically self-detected parameter. And I utilize the method developed by Yuille and Naonori to improve the arithmetic for the system stability.Through research and experiments, I compared several Pattern Recognise methods with different operation speed, recognise rate. And I also research some ways of threhold adjusting. All the contents which can be identified include the printed material, the manuscript numbers, all the bank names and alphabets.I have researched it in all kinds of conditions with different paper material, paper surface, different water mark depth and different table forms. The images should be processed before verifying, especially to remove the frames of the manuscript amounts. As for this action, I adopted Hough change, and erase the parallel table line by the exchange between polar coordinates and ordinary coordinates.