Research On Printed Tibetan Character Recognition Technology Based On Modified Rough Grid

Character recognition is set for pattern recognition,image processing and word processing technology with a new technology and is an important research area of parrern recognition and artificial intelligence. Today,after years of exploration and practice,Western and Chinese printed text recognition is moving toward a practice use. Because of the complexity in Tibetan characters structure and overfull similar characters, Tibetan character recognition is always considered as a challenging problem.Based on the existing printed recognition technology,this paper studies the preprocessing, feature extraction and recognition and classification algorithms. A novel feature extraction algorithm to printed Tibetan characters based on modified rough grid is developed in this paper.Major work completed is as follows:(1) Firstly,the preprocessing of Tibetan character image is discussed in the paper.The different methods of binarization, smoothing, lines and words extraction and normalization are studied,and the different ways are compared and the paper chooses the most suitable for dealing with printed Tibetan characters later in the pretreatment methods. Characters with the same size and the less noise are got after preprocessing.(2) Three feature extraction algorithms:image projection method,directional line element method and fractal moment method are introduced and compared here. In view of the previous lack of three methods, a novel feature extraction algorithm to printed Tibetan characters based on modified rough grid is proposed in this part. The experimental results show that the extracted feature by using the new method can effectively decrease the influence that caused by the decline in the rate of recognition part due to the change of pixel's position in the image and the method overcomes high probability of misrecognition due to numerous similitude Tibetan characters.(3) The extraction of original features with relativity and redundancy reduces the recognition rate and recognition speed, so feature selection is essential after extracting original Tibetan character feature. This paper uses principal component analysis (PCA) to select features and carris out simulation experiments.(4) The design of classification is the more important issue of pattern recognition. In the classifier design process, different feature selection methods and different classification algorithms and different measurement methods are used. This paper uses BP neural network to classify. At last, Tibetan feature extraction, selection and identification are done based on modified rough in this paper and some experimentals of Tibetan character recognition are conducted, and are compared with the directional line element method. The experimental results show that the modified rough grid algorithm to deal with the printed Tibetan characters has the higher recognition rate and recognition speed.