Research On Handwritten Chinese Character Recognition Based On Computational Intelligence

Computational intelligence is a new and developing computing method,which has many advantages in its characters, such as intelligence, parallel processing and self-adaptive ability. It provides new effective avenues of dealing with many complicated problems. This paper mainly discusses two novel intelligent optimized algorithms——Particle swarm optimization algorithm and artificial immunization system. The neural network models trained by the two methods respectively are provided. Then this models are applied to recognize the off-line hand-written Chinese characters, and confirmed the effectiveness of the two intelligent optimized algorithms through many tests. The main researches are as follows:(1) Based on the discussion of general problems in hand-written Chinese characters recognition, a new feature extraction method adapted to the particularity of Chinese characters is provided, which includes character preprocessing, image thinning, extracting direction features and so on. The accuracy of Chinese characters recognition is improved and the validity of the intelligent algorithms is guaranteed by using these kinds of feature extraction.(2) In order to overcome the shortcomings of BP algorithm which used to train the neural network, such as the slow speed, the big errors, a novel training method based on Particle swarm optimization algorithm (PSO) is presented. The important parameters in PSO are analyzed and improved, and a model of neural network based on improved PSO algorithm is set up. Theory analysis and the experiment result show that the improved PSO in neural network training can achieve high precision and get very high convergence speed.(3) The thought of immunity is introduced into the hand-written Chinese characters recognition. According to object problem, all elements in immunity system are established, genetic algorithm is applied in immunity system, and some correlative mathematics expressions are defined. So a neural network model based on immunity system is designed, which is very effective in Chinese characters recognition. Furthermore, the improved PSO and AIS algorithm are compared with that of traditional BP, and the result shows that the intelligent computing is superior to the traditional methods at precision and convergence speed in neural network training.