基于Contourlet域HMT模型的多尺度图像分割

MULTI-SCALE IMAGE SEGMENTATION BASED ON CONTOURLET-DOMAIN HIDDEN MARKOV TREES MODEL

基于Contourlet系数分布统计特性,结合隐马尔可夫树(HMT)模型和贝叶斯准则提出一种新的图像分割算法.为了更有效保持Contourlet域不同尺度间的信息,提出一种新的加权邻域背景模型,给出了基于高斯混合模型的象素级分割算法和基于新的背景模型的多尺度融合算法.分别选择合成纹理图像、航拍图像和SAR图像进行实验,并与小波域HMTseg方法进行比较以说明算法的有效性.对合成纹理图像给出错分概率作为评价参数.实验结果表明本文方法不但在边缘信息和方向信息保持上有明显改进,而且错分概率明显降低,对真实图像得到了理想的分割效果.

Based on the statistics characteristics of contourlet coefficients, a new multi-scale image segmentation method （CHMTseg） combining Contourlet domain hidden Markov trees model with multiscale Bayesian approaches was presented. A novel weighted neighborhood model was given for preserving more inner-scale information in Contourlet domain. The pixel level segmentation based on Gauss mixture model and the multiscale fusion method based on the new contextual model were provided. In experiments, synthetic mosaic image, aerial image and SAR image were selected to evaluate the performance of the method, and the segmentation results were compared with wavelet domain HMTseg method. For synthetic mosaic texture image, miss classed probability was given as the evaluation of segmentation results. Experiment results show that the method not only has better performance in edges and anisotropy information detection but has lower missed classed probability, and it can achieve satisfied segmentation results for real images.