A LBP-and-PCA-based Recognition System For Off-line Handwritten Tibetan Character

With the continuous development of the pattern recognition technology, the understanding of the Tibetan character image has increasingly become a research hotspot, and many scholars have been attracted by the off-line handwritten Tibetan character recognition. At present, there are few research results of off-line handwritten Tibetan character recognition, and the author would like to make a further research in this field. The samples have been selected from THCDB, a newly constructed database of handwritten Tibetan character samples. These samples have been preprocessed, so this paper mainly focuses on the feature extraction methods and classification method.The main purpose of feature extraction is to remove noise and redundant information, and retain the useful information so as to improve the effectiveness of features and make them more concise. Firstly, the local binary pattern(LBP) histogram has been proposed as Tibetan character features. The recognition rate of this feature is better, but the disadvantage is that the dimensionality of the feature space is larger and the recognition speed is slower. Therefore, principal component analysis(PCA) has been employed to reduce the feature space dimension of LBP histogram feature. The experimental result shows that when the dimension of the newly proposed feature space is 196, the recognition rate is optimal.Classifier design is an integral part of a recognition system. KNN(K-nearest neighbor) is a classifier that is simple in principle and easy to operate. Therefore, KNN is adopted firstly to recognize the Tibetan characters in this paper. The optimal recognition rate of the 30 Tibetan consonants is 96.04%. In order to further improve the recognition rate of the Tibetan character, this paper further uses the support vector machine(SVM) classifier. The recognition rate can be affected by the different kernel functions of support vector machine and the experimental results show that the optimal recognition rate 97.93% can be reached as we use radial basis function.In short, for handwritten Tibetan character recognition, a better recognition result had been reached as we adopt the local binary pattern histogram and principal component analysis as features, and radial basis function support vector machine as classifier.