Image fusion using a contourlet HMT model

In this paper hidden Markov tree (HMT) based image fusion methods are investigated. Considering the failure of wavelet in representing the geometry of image edges in dimension 2, here a new contourlet HMT model for image fusion is proposed. Because the CHMT model efficiently captures all dependencies across scales, space and directions through a tree structured dependence network, it can give more accurate description of images. Moreover, the CHMT has a simple tree structure with fewer parameters than wavelet HMT (WHMT), which enables efficient training using the expectation maximization (EM) algorithm. Inputting the contourlet coefficients of source images to train the CHMT model, we can get the edge probability density functions. Local inner-product fusion rule is performed on the high-frequency directional sub-bands, which is acquired by the product of the high-frequency directional coefficients by the edge probability density function of CHMT. The low-frequency sub-bands are compared to preserve the coefficients whose module are minimum. The experiment results show the superiority of the proposed image fusion method to WHMT and contourlets, both in image clarity, implementation speed, standard deviation, average gradient and average cross entropy.