Abstract: The mistaking samples gotten by the interactive way greatly decrease the performance of the SVM-based segmentation method. In this paper, a segmentation method based on a 2 level SVM was proposed to reduce the negative effects of the mistaking samples. The first level SVM filtered out the mistaking samples, and the second level SVM segmented the images. A SVM-based filtration method was also adopted. The experiments illustrated that the proposed method could get better result.

Key words: Support Vector Machines, Sample filtration, Image segmentation

摘要： 基于支持向量的图像分割方法一般使用交互方式获取的训练样本，不可避免的在训练样本中引入歧义样本。这些歧义样本严重影响了基于支持向量机图像分割方法的性能。本文提出一种先对训练样本进行筛选，再进行分类（分割）的支持向量图像分割方法；并给出了一种基于支持向量机的样本筛选方法，可有效的降低歧义样本的影响。实验表明，经样本筛选的SVM分割方法有更好的分割性能。

关键词: 支持向量机, 样本筛选, 图像分割