Multi-focus Image Fusion based on Multi-scale Focus Measures and Generalized Random Walk

Multi-focus image fusion aims to produce an all-in-focus image by integrating a series of partially focused images of the same scene. A small defocused (focused) region is usually encompassed by a large focused (defocused) region in the partially focused image, however, many state-of-the-art fusion methods cannot correctly distinguish this small region. To solve this problem, we propose a novel multi-focus image fusion algorithm based on multi-scale focus measures and generalized random walk (GRW) in this paper. Firstly, the multi-scale decision maps are obtained with multi-scale focus measures. Then, multi-scale guided filters are used to make the decision maps accurately align the boundaries between focused and defocused regions. Next, the GRW is introduced to effectively combine the advantages of the decision maps in different scales. As a result, our method can effectively distinguish the small defocused (focused) regions encompassed by large focused (defocused) regions, and the boundaries can also be aligned accurately. Experimental results demonstrate that our method can achieve a superior performance compared with other fusion methods in both subjective and objective assessments.