摘要
提出了一种基于小波域高斯—马尔可夫随机场(GMRF)模型的无监督纹理图像分割算法。该算法首先利用纹理的小波特性并结合模糊C—均值聚类方法完成纹理在最小分辨率层的初始分类,接着逐层应用同步进行参数估计和像素标签的近似最大后验分割算法,得到原始图像的第一次完整分割。为了进一步提高分割效果,对每个像素邻域内的标签作统计,利用最大值原则,从而获得满意的分割结果。实验证明此算法与基于高斯金字塔GMRF模型的算法相比,分割结果有了很大的提高。
An unsupervised texture image segmentation method based on Gauss-Markov random Ileitis (GMRF) is proposed.Some wavelet features combined with fuzzy c-mean clustering algorithm is used to accomplish the initial classification in the most coarest resolution level.The approximate maximum a posteriori segmentatlon,which estimates parameters and labels pixels simultaneously,is applied layer by layer so as to complete the first integrated segmentation for the original image.ln order to acquire a better result,we statistics the quantities of labels in the neighborhood of every pixel.According to maximum principle,a satisfying output is received.Compared with the method based on Gausspyramid markov random fields,experiments have proved that our method is superior.
出处
《计算机工程与应用》
CSCD
北大核心
2005年第22期39-42,共4页
Computer Engineering and Applications
关键词
小波分解
高斯-马尔可夫随机场
无监督
近似最大后验概率
wavelet decomposition,Gauss-Markov random fields,unsupervised,approximate maximum a posteriori