Based On The Dsp Notes Character Recognition Technology Research

As the main medium of the modern commodity exchange, banknotes have brought great convenience to our daily economy life, but at the same time, the management and statistical work of the banknotes has become increasingly complex, the traditional statistical methods have been unable to adapt to the changes in this phenomenon. To solve this main problem for banknotes number recognition and identification in this paper, support vector method is applied to identify the characters of banknotes number, and all the recognition algorithms are ported to hardware platforms which use the TMS320DM642as the core processor to rapidly and accurately identify the characters of banknotes number. On banknotes, for the realization of rapid and accurate identification, this dissertation mainly research from these aspects of hardware platforms, algorithm optimization and system software:1. Implement a hardware platform with DSP as the core processing chip and test it by the library functions provided by TI company, as well as other TI’s resources, such as CSL for debugging and so on. Also optimizing the configuration registers to lay a solid foundation on running these algorithms.2. Analyze various pre-processing algorithm, the feature extraction algorithm and the recognition algorithm in theory and simulate these algorithms in MATLAB. Combining with the characteristics of characters of the banknotes number, analyze their advantages and disadvantages in performance and select graying algorithm, Binary image of the character region by global threshold value,a single character segmentation for character pre-processing, and extraction of its characteristics and use standard support vector algorithm to identify the character. After simulating the entire algorithm in MATLAB, these algorithms were transplanted to the hardware platform to process and identify these characters.3. Analysis of the mathematical principles of SVM and do some research on the parameters which the SVM require and select one-vs-one training algorithm, SMO training algorithm and radial basis kernel function SVM parameters and so on. Set up classifiers needed by the formation of the whole SVM system.4. Combined with the hardware platform and algorithm characteristics, write the entire system software.