Multi-focus Image Fusion by Local Optimization over Sliding Windows

This article presents a technique to solve the problem of multi-focus image fusion. This technique is based on the maximization of a linear function with spatial coherence constraints. The final fused image is computed as the sum of the source images using a segmentation map. We can compute the segmentation map using the Simplex method, where the objective function includes one variable associated with each pixel. The Simplex method requires a huge amount of memory resources to produce it. We present an algorithm called CPW-S, which uses some strategies to solve the problem in a context with fewer variables; images are split into regions, thus reducing the computational effort. We present results for two pairs of synthetic images in order to quantify the results, obtaining more than \(98\%\) of pixel accuracy for the segmentation map. We also present results for several pairs of real images (widely used in the literature) and a triad of multi-focus images. The resulting fused images are qualitatively good for all the real images included in the experiments.