Technological research on an application of neural network in character recognition
Ping Li, Xiaofeng Mu · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1998
The artificial neural network Abstract simulates the principle of human neural element movement. It has many abilities such as study independently, adaptation itself , association,reasoning from analogy and generalization. The application of the artificial neural network in character recognition. The two procedurcs of character recognition ,i. e. ,the procedure of learning and the procedure of recognition ,were comprehensively described in this paper. The procedure of learning, that is, the procedure of training, the characteristic parameters produce the standard sample, and characters are recognized by using the standard sample, it can make continuously adaptation itself correction in use by the results of recognition. Because, of adopting the improved BP algorithm, reasonably selecting initial values, the adjusting method of learning rate, putting in momentum and introducing the simulated annealing method, the problems of the slow convergence speed of the original algorithm, bogging the local minimum and training paralysis was solved to improve the recognizing rate and obtained satisfactory results.