A robust approach for multi-sensor medical image fusion

Abstract

This paper demonstrates a robust method to the multi-sensor medical image fusion (MIF) obtained using the non-subsampled contourlet transform (NSCT), generalized Gaussian density (GGD), and Kullback-Leibler divergence (KLD). The popularly used average-maximum fusion rule is able to capture the local information only. However, it is unable to capture the global information. Hence, a global-to-local rule is suggested here. First of all, NSCT is used to separate the low and high-frequency sub-bands from the given source images. The heavy-tailed phenomenon of high-frequency sub-bands is modeled by GGD. The KLD for two source images is obtained by using GGD of them. This is used to describe the global information between two sub-bands. Finally, according to the asymmetry of the KLD, the fused global information is selected. The proposed method is able to overcome the different issues arise in the state-of-the-art methods such as reduction in contrast, loss of fine details, etc. The proposed algorithm is executed on the various datasets and its results show that the proposed algorithm provides better results than the existing MIF algorithms.