Multi-focus image fusion based on smooth and iteratively restore filter

Abstract

Image fusion is the process of combining several images with different focus settings into a completely in-focus image. However, many state-of-the-art fusion methods cannot well preserve all the significant features of the source images to obtain an all-in-focus image. In this paper, an improved smooth and iteratively restore (SIR) filter is proposed to deal with the problem. The SIR filter can well smooth noise while retaining details of edges. First, SIR filtering is applied to decompose source images into base and detail layers. Second, saliency maps of the different layers of the sources are computed by a proposed salient feature filter. Third, pixel-wise maxima of the saliency maps are used to construct the binary decision maps for both source images. Then with the binary decision maps we fuse the base and detail layers respectively to yield the fused base and fused detail, which are then recombined to produce the final fused image. Spatial consistency is inherently guaranteed by this process. Tests on sets of grayscale and colour multi-focus images demonstrate that the proposed method achieves better performance than existing methods in terms of both subjective and objective evaluations.