A quantitative performance evaluation index for image fusion: Normalized Perception Mutual Information

Information entropy based criteria are analyzed and the Normalized Mutual Information(NMI) that is presented in the field of image registration is revised to Normalized Mutual Information Entropy (NMIE) to meet the need of the evaluation of image fusion algorithms. Then, through analysis to NMIE and some human perception based criteria, and by analyzing the essence of image fusion techniques systematically, a new index, Normalized Perception Mutual Information (NPMI), is defined in view of information transmission as well as edge preservation, and is used to evaluate the performance of image fusion algorithms. The experiments are done to three groups of images, namely, the remote sensing images corrupted by noises, the multifocus images and the medical images obtained by CT and MRI, respectively. Compared with other indices including the root mean square error (RMSE), space frequncy (SF), space visibility (SV), entropy, the collective cross entropy (CCE), information deviation (ID), and the edge information preservation value (EIPV), etc., NPMI is shown to be the only one that is effective in all the cases in the evaluation of the performances of the fused images or the image fusion algorithms, which illustrates the feasibility and effectiveness of the presented algorithm.