Character Recognition Algorithms And Realization On Scanned Cheque Images

With the rapid development of computer technology, information technology has brought many changes to human lives and greatly promoted the development of human society. Characters as a carrier of social civilization play an important role in the field of information technology. Optical Character Recognition, short for OCR, which can mitigate human workload and save working time, have been widely used in many applications, especially in the field of finance, communications, journalism and publishing. Therefore, this thesis mainly focuses on the algorithms of printed character recognition on the bank cheques.Firstly, this thesis conducted pre-processing for characters recognition of cheque image, in order to facilitate the post processing and improve the recognition effect of OCR. And then it studied some common image pre-processing methods, used median filter for image restoration and remove noise, and selected OSTU algorithm and Radon transform method for binarization and tilt correction of cheque image.Then, with the cheque image layout analysis and projection of the character area, this thesis further extracted the character region of cheque image and operated thinning and segmentation of the printed characters. Subsequently, this thesis has studied and simulated the character feature extraction and recognition algorithms. It has normalized the segmented characters, analyzed some common methods of character feature extraction, and finally selected 13-feature extraction method for character feature extraction. Afterwards, it has researched different features of the characters and studied two kinds of common methods for character recognition, the character recognition method based on template matching and the Bayes classifier method based on binary data. Then the thesis simulated these two character recognition methods after done character feature extraction of the cheque image, and analyzed the simulation result. The recognition rate was not high and the result was not satisfied.Finally, after classification studied of algorithms for OCR, combined with the identification process of printed characters of cheque image and based on the neural network model, this thesis deeply studied model structure, learning style and learning rules of the neural network. In the end, according to the actual cheque images, this thesis designed the printed character recognition system for cheque images based on BP neural network, then trained the system by many samples and conducted figure amount recognition of real cheques, and analyzed the recognition results. The recognition rate was high and the result was satisfied.