A Multi-Classification Method of Liver Pathology Images Based on Sparse Multi-Scale Local Binary Pattern-Local Directional Pattern

摘要

For traditional Local Binary Pattern (LBP), the amount of calculation will increase with the increasing of the template radius. The single radius of template can not collect the information of different regions. So in this paper, a sparse multi-scale LBP and Local Directional Pattern (LDP) method is proposed to classify liver pathology images. Sparse means sampling selectively, and multi-scale means sampling points of different radius. The proposed sampling method samples points of different regions, which makes the information more representative, and the computational efficiency is not decreased for sampling selectively. Higher Order Local Autocorrelation Coefficients (HLAC) feature and Gray Level Co-occurrence Matrix (GLCM) feature are also introduced into this paper for its effectiveness of classification. Kernel Principal Component Analysis (KPCA) is used to select less number of important features, which will reduce the dimension of the feature vector, and the classification accuracy will also be improved for the removal of redundant features. The grid parameter optimization is introduced to find the best parameters of SVM classifier. Lastly, optimized SVM is used to classify multi-class liver pathology images. The experiment result shows the proposed method can classify multi-class liver pathology images more accurately than the original method.