基于Contourlet域隐马尔可夫树模型的图像融合

摘要

基于Contourlet域的隐马尔可夫树模型能反映不同尺度系数之间、不同方向系数之间的相关性,基于此,提出了一种基于Contourlet域隐马尔可夫树模型的图像融合算法.对源图像进行Contourlet变换,并针对高频子带系数建模并训练得到每一系数的后验概率；利用该后验概率指导高频系数融合的规则,对边缘和背景区域进行不同的融合处理,以尽可能保留原始图像的重要特征；进行Contourlet反变换得到最终融合结果.针对多聚焦图像进行了融合实验,采用联合熵、熵、相关系数、清晰度等指标对融合效果进行评价,实验表明了该算法优于基于Contourlet域的常规融合算法以及小波域隐马尔可夫树融合算法.%Contourlet-domain hidden Markov tree model can reflect the coefficients' relevance of different scales and directions, a image fusion algorithm based on it is proposed.Source images are processed by Contourlet transform, low frequency and high frequency subband coefficients are obtained.The high frequency subband coefficients are modeled using hidden Markov tree and the model is trained using EM algorithm to get the posterior probability of the coefficients.Using the posterior probability to guide the fusion rules' design of high frequency subband coefficients, it fuses the coefficients differently according to edge or background region.The fusion coefficients are inversed to get the final fusion result.Multi-focus images are taken to do fusing experiment, and the results are evaluated by joint entropy, entropy, correlation coefficient and clarity. The experiment shows that the proposed algorithm is better than traditional fusion algorithms based on Contourlet and wavelet-domain hidden Markov tree model.