Application And Research Of Pattern Recognition In Character Recognition

Offline handwritten character recognition as an important component in the fieldof pattern recognition, have been playing a pivotal role, with a very high significance.Throughout the history of text recognition researching, the study of writing both inchina and abroad for Chinese, English, Japanese, Korean, and the others endless haseffectively yielded gratifying achievements, and mature application. But little for therelatively small and dispersed crowd, especially the Arab text.Having carefully studied the popular recognition method of advantage writing,combined with the characteristics of Arabic text, introducing BP neural networkclassification policies, and put forward two improvements for its shortcomings.firstly, This Paper has analysis the reasons for offline Arab text recognition cannotobtained ideal effect and it’s research status; secondly, it described and analysed thekey technology in offline text recognition, and showed the image preprocessing andextraction features method using in this article; again, it introduced the principle ofneural network classification algorithm and the main application in detail, at the sametime it analysed the disadvantages and insufficient; shortly thereafter, this papergave two species of improved classification algorithm, respectively is PSO-BPalgorithm and ELM algorithm, and did some analysis on their performance, PSO-BPalgorithms significantly shortens the training time, play a good role on optimization ofBP networks structure, ELM gives a innovation on the mechanism of BP training,this paper also demonstrated its rapid and efficient features; Finally,combaine theIFN/ENIT handwritten Arabic text databases constructed a library, and then usingthese modified classifiers for training and testing. By experiments to verify theeffectiveness and efficiency of the improved methods, recognition rate of PSO-BPalgorithm classifier can reach92.3%,and ELM classifier recognition rate can reach93.1%.