Research On Image Classification Based On Support Vector Machine

With the development of digital equipment, network and multimedia technology, Digital images that are created in people's working, learning and daily life appear explosive growth mode.It is one of hot research topics that how to organize the mass image data effectively and how to classification and retrieval the digital image with the image low level characteristics.Support Vector Machine(SVM) is a statistic learning method based on less samples proposed in recent years and can well resolve such practical problems as nonlinearity,high dimension and local minima.A large number of experiments have shown that SVM has not only simpler structure, but also better performance, especially its better generalization ability, so it is considered to be an efficient classification algorithm. Therefore, SVM-based Image Classification has been the important theory and technology of Image Classification.The main content of this paper are listed as follows:First, because only one kind of image can not describe image content completely, a multi-feature integration combing color texture and spatial feature is extracted and used as input training vectors for SVM.Second, the influences of the error penalty parameter C and the kernel parameter s on SVM's generalization ability are studied. On the basis of analyzing the parameter's influence on the classifier'classification ability, a new approach of Parameters Selection of SVM Based on Genetic Algorithm was proposed. The results demonstrate that the algorithm can get the SVM with the best recognition accuracy and simple structure.Finally, SVM for Multi-class Classification is discussed. Several methods have been proposed including "one-against-all" ,"one-against-one" , DAGSVMS, Classific- ation method of mufti-class SVM based on binary tree(BT-SVMS), and so on. And their pluses and minuses and performances are compared.Next, this paper proposes a classification algorithm of multi-class SVM based on GA and KNN. GA is used to create optimal or near-optimal classification tree automatically, while in the process of integration into the KNN classification algorithm to improve classification accuracy. Experiments show that the proposed method is more precise than the traditional BT-SVMS method, and has more reasonable classification tree structure.