Offline Handwritten Chinese Character Recognition Based On Sparse Representation

Offline handwritten Chinese characters recognition is a research hotspot in the present field of pattern recognition, and has broad application prospects. However, because of more types and more complex structure of Chinese characters, and in addition, the serious deformation of character shape caused by different writing stvles of different persons, offline handwritten Chinese characters recognition is always one of difficult problems in the present field of Chinese characters recognition.Some methods of the image processing can be also applied to the Chinese character processing because the Chinese characters possess the characteristic of images. In this paper, we study and summarize carefully the offline handwritten Chinese character recognition algorithm and the previous works by analyzing and investigating the analogy between images and Chinese characters. Based on that, we apply the Group Lasso algorithm, which is used to the image recognition, to the offline handwritten Chinese character recognition, carry out a lot of experimental research, put forward an offline handwritten Chinese character recognition method on basis of the sparse representation, and obtain good effects. The paper is arranged as follows:Firstly, we analyze each procedure of the character image preprocessing, and point out its effect on improving the follow-up recognition process and its importance. We analyze two types of existing feature extraction methods:statistical feature extraction method and structure feature extraction method for their applied field, and investigate several classical classifier algorithm, respectively, based on the distance. Fuzzv identification, artificial neural network and support vector machine. Summarize the respective advantages and disadvantages of these algorithm, and lay a foundation in order to find a new recognition algorithm.Secondly, noticing the case that the recognition efficiency that the present several offline handwritten Chinese characters recognize classical algorithm is still not high, based on the compression sensor theory, and according to such a feature that the sparse representation possesses good distinction, we propose a scheme to identify sparse representation for handwritten Chinese characters, and give a specific algorithm-Group Lasso algorithmFinally, contraposing sparse representation Group Lasso algorithm, using the Chinese characters in SCUT-IRAC handwritten Chinese character database as a sample, in the MATLAB R2009B software environment, we adopt the methods in this paper and other classical algorithm, and perform the simulation experiments. The results show that this method successfully improves handwritten Chinese character recognition rate and anti-interference capability and possesses high promotion of high popularization value.