Research On Texture Classification Based On ICA And SVM

Texture is an important attribute in the image, which provides substantial information for the recognition and interpretation of this image. The research problems of the image include texture perception, texture analysis, texture synthesis and so on. Texture analysis is a main and useful area of study in pattern recognition. Texture classification is an important problem in the texture analysis, texture feature extraction and classification are crucial in the texture classification.In recent years, Independent Component Analysis(ICA) has been highlighted in statistic signal proposing. The basic principle of ICA is to find the independent hidden information through analyzing the high-order statistic relation of observed data, and achieve the goal of getting rid of the redundant high order component and extracting the independent source data. ICA has the characteristics by which we can recover the independent hidden source from observed data in the condition that there is no information about the source data and the mixing mechanism.This advantage makes ICA widely applied in image feature extraction, pattern recognition and so on.Support Vector Machines(SVM) is based on statistical learning theory(STL). STL is a theory that specialized in machine learning with finite samples. SVM is considered as a candidate to replace neural networks and other traditional classification methods for its good performance and high generalization ability.A novel classification method for texture images was proposed by combining independent component analysis(ICA)with support vector machines (SVM).The basis functions were extracted by ICA.The feature vector was generated by computing mean and standard deviation from convolution of texture with basis functions. This feature vector was used for training and testing the support vector machines classifier. The experimental set up consists of four and ten texture images from the Brodatz image database. The results of classification experiment show that the classification accuracy of the presented method is superior to that of other methods under the condition of the limited training samples, with the validity and better generalization ability. Experimental results show that the proposed algorithm can effectively improve the accuracy of the image classification.