The Research Of The Image Classification Method Based On Support Vector Machine

Now in the field of image retrieval, the multimedia image retrieval of semantic-oriented is the fashion trends. But how solve the problem of the "semantic gap" has been troubled people. More and more people began to study the support vector machine (SVM), since the SVM show a good performance in the field of machine learning. Then the SVM is applied to the field of the image retrieval, through the way to improve the accuracy of the image classification, and solve the problem of the "semantic gap". On the basis of the depth study of the basic principles of SVM, and study these multiclass classification algorithms of SVM which published by others, the paper advance an idea of improved SVM classification algorithm, which based on the distance and normality binary tree.Now to run a image classification algorithm is very time-consuming on a single computer, because the algorithm has a complex calculation. With the advent of cloud computing platform, the computing power of the computer cluster can improve the algorithm calculations speed of the image classification. This article attempts to improve the computing speed of the algorithm, then combine the improved algorithm with the Hadoop platform.The main works of this paper are as follows:(1) This paper briefly introduces the background、significance and status of current image classification retrieval、support vector machine and cloud computing;(2) This paper detailed introduces the related theory of SVM, the multi-class classification algorithm of SVM. And all the kinds of algorithm advantages and disadvantages are analyzed. We found that the binary tree SVM class classification algorithm has more obvious advantages;(3) This paper briefly introduces the relative concepts of cloud computing, and then laid special stress on making a detailed description for the Hadoop, which is the cloud computing open source platform. Especially the reading and writing strategies of the Hadoop, the job flow of the MapReduce, this paper has a detailed explanation;(4) This paper presents the improved binary tree SVM classification algorithm, and then through the experiments show the improved algorithm has better classification accuracy and good classification speed;(5) This paper combine the improved SVM algorithm with the MapReduce, and finally proved that good by the experiment, this combination can helps to reduce the calculation time of the image classification complex algorithms.