MGFuse: An Infrared and Visible Image Fusion Algorithm Based on Multiscale Decomposition Optimization and Gradient-Weighted Local Energy

Existing image fusion algorithms have difficulty in effectively preserving valuable target features in infrared and visible images, which easily introduces blurry edges and unremarkable notable targets during their fusion process. We propose the MGFuse algorithm as a solution to this problem, which is a novel fusion algorithm that utilizes multiscale decomposition optimization and gradient-weighted local energy. Initially, non-subsampled shearlet transform (NSST) is applied to partition both the infrared and visible images into several high-frequencies and low-frequencies components. Subsequently, the acquired low frequencies continue to be decomposed via the proposed optimization function to get base layers and texture layers, which can optimize the quality of image edges and preserve fine-grained details, respectively. In addition, we have formulated an intrinsic attribute-based energy (IAE) fusion scheme to merge the two base layers. The texture layers and high-frequencies are extracted by gradient-weighted local energy (GE) operator based on structure tensor, which is employed to construct the fusion strategy for these parts. At last, the acquired texture and base parts are linearly combined to get the integrated low-frequency layer on which the final image is acquired using inverse NSST. Numerous experimental observations demonstrate that our MGFuse algorithm achieves superior fusion capability than the reference nine advanced algorithms in both qualitative and quantitative assessment, and robustness to noisy images with different noise levels.