Research Of Offline Handwritten Character Recognition Technology

The problem solved by Offline handwritten character recognition technology ishow to automatically identify handwritten characters in the image by the computer inthe offline state, and it meets the requirements of the fast text input. The technique isone of the hot research in the field of pattern recognition, it involves a number of keytechnologies which it contains image preprocessing, feature extraction and the designof classification and identifying and it becomes it be a challenging research of thissubject. In addition, because different characters have different recognition methods,the researching work on offline handwritten Chinese characters recognition has theimportant theoretical significance and reference value for the recognition of minoritylanguage in Xinjiang.The current methods of character recognition and classification always existsome defects, such as template matching has anti-jamming ability. Support vectormachine learning is only suitable for small sample classification. HMM for characteris not easy to build and has the poor robustness. BP network has strong adaptabilityand self-learning ability. But the difficulty of the network structure to determine andthe selection of the training parameters easily lead to training time, slow convergencerate, and falling into local minima.The paper has improved13-points feature extraction method and the presentationcapabilities of the method. This paper determines the number of neurons in each layerof the network through the experiments; adding a constant in S-type driving functionaccelerates the convergence speed of the network. In selecting the training parameters,the learning method is proposed that the Levenberg-Marquardt algorithm improvesconnection weights and thresholds of the network.Based on the previous work of handwritten numeral character recognitionsystem, the paper builds offline handwritten character recognition system. The system uses the improved LM-BP network as classifiers and achieves theclassification and recognition of handwritten Chinese characters. Through theoperation commissioning, training and recognition of the system, the results show:under the same condition of training target, the training time of the improved LM-BPnetwork is shortened by95%than the BP network, and the convergence speed hasalso been improved. The recognition rate of the classifier is improved by6.4%onaverage. The results confirm that the offline handwritten character recognitionsystem is efficient and feasible.