基于空间分布支持向量机的图像分割

Image segmentation based on SVM using spatial patterns

利用模糊聚类与支持向量机结合的方法,将图像的空间分布信息作为支持向量机的特征分量,并用模糊聚类获得的分类结果作为支持向量机的初始训练样本对图像的所有像素点进行分类,同一类中的像素点形成一个分割区域,以此获得图像分割.实验表明,该方法获得的图像分割效果较好,在一定程度上解决了特征维数过大所导致的维数灾难问题.

We propose a new hybrid methods for image segmentation that base on support vector maching（SVM） combined with C-mean fuzzy clustering.This method spatial pattern information is used as component characteristics of the SVM,and the classification results from fuzzy clustering are used as the initial samples of the SVM.Then the pixels of the image are classified by SVM and the pixels in the same class form a segmental region.The experimental results show that the new methods combing fuzzy clustering and SVM can get better results and the error ratio caused by the segmentation is decreased.