Contourlet变换的图像分割

摘要

针对小波域隐马尔可夫树模型分割的图像结果容易产生方向边缘成分模糊和奇异性扩散现象,根据Contourlet变换可以充分捕捉图像中高维奇异性,提出了一种基于Contourlet变换域的新的多尺度图像分割算法.该算法通过Contourlet域隐马尔科夫树模型获得各尺度上的初始分割,采用自适应的上下文结构对分割后的图像进行多尺度间的融合.对合成纹理图像和航拍图像进行分割实验仿真,并与基于小波域隐马尔可夫树模型的图像分割方法进行比较,区域一致性和边缘准确性得到改善,得到更为理想的分割效果,对合成图像降低了错分概率.%By using Wavelet Domain Hidden Markov Tree HMT model, the image segmentation resulted in dim edge components and phenomena with singularity and diffusion. Because the singularity in the image of high dimension was fully captured by the Contourlet transformation, the paper presents a new algorithm of multi-scale image segmentation based on the Contourlet transform of domain Hidden Markov Tree model. This algorithm was initially through image Contourlet transform to obtain transform coefficients, to train Contourlet domain Hidden Markov Tree Model, to calculate multi-scale likelihood function and obtain the reliable initial segmentation based on the maximum likelihood estimation (ML) formula, and finally to use adaptive context measurement integrated methods to get ultimate segmentation image. Through the simulation experiment of synthetic texture image and aerial image, compared image segmentation methods based on wavelet domain Hidden Markov Tree model, either visual effects or evaluation parameters illustrate the effectiveness of the algorithm mentioned in this paper, obtain better regional consistency and edge accuracy, and reducing the error probability of synthetic images.