Abstract:Image segmentation is an important topic in image understanding and computer vision. When support vector machine(SVM) is used for image segmentation, the design of its kernel and selection of the parameters directly affect the segmentation effect. Considering the problem that SVM based on single kernel could not keep the balance between the segmentation accuracy and generalization performance, an image segmentation algorithm using optimized multi-kernel SVM(OMKSVM) based on K-means clustering was proposed in this paper. According to the multi-kernel learning theory, the new multi-kernel is constructed by a linear combination of single kernels. Firstly, the K-means clustering algorithm was applied to obtain the training samples for MKSVM automatically. Then color and texture features were extracted from the image as attributes of training samples of MKSVM, Particle Swarm Optimization(PSO) algorithm was employed to optimize the kernel parameters, the weight coefficient and the punishment coefficient of SVM simultaneously. Finally the OMKSVM was obtained to segment image. Three groups of complex color image were selected to verify the correctness of the proposed method. The results demonstrate that our method can segment the color images effectively, and has stronger generalization ability comparing with the single kernel SVM-based method.