Multi-scale fusion algorithm comparisons: Pyramid, DWT and iterative DWT

An advanced discrete wavelet transform (aDWT) for image fusion was recently proposed and the aDWT algorithm was further optimized with an iterative procedure utilizing quantitative fusion metrics - the image quality index (IQI) and ratio spatial frequency error (rSFe), respectively. In this paper, a modified rSFe metric by considering the first-order and second-order gradients (denoted as rSFe_2nd) is presented. Using rSFe_2nd and normalized mutual information (NMI) as optimization metrics, two new iterative aDWT (aDWTi) algorithms are developed. For comparison purpose, Laplacian pyramid and regular DWT are also implemented. Spatial frequency and entropy are used as quantitative evaluation metrics. For visual examination, three frequently-used image pairs were analyzed and presented. From the experimental results, it is found that (1) the aDWTi-rSFe or aDWTi-rSFe_2nd algorithm is the best; (2) the aDWTi-IQI provides us a smoothed fusion; (3) the aDWTi-NMI algorithm is unsatisfied; and (4) Laplacian pyramid is better than regular DWT but poorer than iterative DWT algorithms.