An efficient adaptive fusion scheme for multifocus images in wavelet domain using statistical properties of neighborhood

In this paper we present a novel fusion rule which can efficiently fuse multifocus images in wavelet domain by taking weighted average of pixels. The weights are adaptively decided using the statistical properties of the neighborhood. The main idea is that the eigen value of unbiased estimate of the covariance matrix of an image block depends on the strength of edges in the block and thus makes a good choice for weight to be given to the pixel, giving more weightage to pixel with sharper neighborhood. The performance of the proposed method have been extensively tested on several pairs of multifocus images and also compared quantitatively with various existing methods with the help of well known parameters including Petrovic and Xydeas image fusion metric. Experimental results show that performance evaluation based on entropy, gradient, contrast or deviation, the criteria widely used for fusion analysis, may not be enough. This work demonstrates that in some cases, these evaluation criteria are not consistent with the ground truth. It also demonstrates that Petrovic and Xydeas image fusion metric is a more appropriate criterion, as it is in correlation with ground truth as well as visual quality in all the tested fused images. The proposed novel fusion rule significantly improves contrast information while preserving edge information. The major achievement of the work is that it significantly increases the quality of the fused image, both visually and in terms of quantitative parameters, especially sharpness with minimum fusion artifacts.