Image segmentation based on Support Vector Machine

In content-based multimedia technologies, video object extraction has received more and more attention. In this paper, Support Vector Machine (SVM) is proposed for image segmentation. The SVM is a learning machine algorithm, can reduce the segmentation error which caused by fast motion of the object. Firstly, frame difference combined with morphology of mathematics is applied to extract the object roughly. Then, the gray value of image pixels and DCT parameters are computed as the characters of the image for training SVM. Finally, a hierarchical decomposed SVM binary decision tree is used for classification. Experimental results show that the algorithm is effective and robust.