Super resolution in multifocus image fusion by focused region extraction techniques

Abstract

Low resolution images in digital photography are due to restrictions placed on the focal depth of the image cameras. To resolve this problem, multi-focus image fusion provides a solution by integrating essential information from multiple focused images captured from the same scene and results in single all-in-focus image with high resolution. Focused region extraction and framing accurate decision map are the two crucial factors associated with the multi-focus image fusion. This thesis aims to develop the image fusion algorithm with Super-Resolution techniques for better extraction of focused regions. In this thesis, a novel base-detail decomposition based multi-focus image fusion method using Anisotropic Guided Filter (AnisGF-MIF) with improved focus measure is developed. The fusion framework is done in two phases. First is learning phase, where the decision maps are formed by measuring Sum of the Bilateral based Modified-Laplacian (SBML) of the source images. Guided Filter is employed to refine the decision map. With the fusion phase, the images are decomposed into base and detail layers using Anisotropic Guided Filter. Both the layers are fused by performing weighted average fusion rule with the obtained decision map and the final fused image is reconstructed by summing up the fused base-detail layers. The algorithm applied to both colour and grayscale datasets and evaluated based on three metrics (1) Information based metrics includes Mutual Information (MI), Entropy (E), Feature Mutual Information based on edge (NE) and gradient (NG), (2) Edge based metric (QG) and (3) Similarity based metrics includes Structural Similarity Index (SSIM) and Correlation Coefficient (CC). On comparison with existing fusion methods, the outcome of the proposed algorithm shows its supremacy in providing better fused images.