提出了一种将超图理论与图象的空-频域特征分析相结合的图象分割方法.该方法是基于图象的多分辨率小波分析及高斯-马尔可夫随机场理论,在抽取一组反映图象局部空间结构信息的特征矢量基础上,根据同一区域的象素具有相似的特征矢量的原则,将图象转换为一个关于特征相似性测度的邻域超图,再利用覆盖-选择算法对该超图进行分割的方法.实验证明,该方法具有较好的稳定性和适应性,尤其对于一些信噪比较低的图象,也具有良好的分割效果.

In this paper, an approach for image segmentation incorporating the theory of hypergraph with the spatial-frequency features of image is proposed. Based on the multiresolution wavelet analysis and the theory of Gaussian-Markov random field, the features characterizing the local spatial structure of image are extracted, a neighborhood hypergraph with respect to the similarities of the features is constructed, segmentation is implemented via the adapted Covering-Selection algorithm.The adaptability and reliability have been tested through simulations. The approach is effective for segmentation of images with low SNR.