Image Fusion Based on Saliency Matrix and Edge Decision Matrix

An image fusion algorithm based on a saliency matrix and an edge decision matrix is proposed to address the limitations of traditional infrared image fusion algorithms in maintaining the thermal characteristics of infrared images and the texture details of visible images. The method begins by utilizing an improved maximum symmetric surround saliency detection algorithm with fast guided filtering to extract the saliency matrix of the infrared image. Concurrently, the Canny operator is applied to the homomorphically filtered visible image to perform edge detection, resulting in the edge decision matrix. Subsequently, the nonsubsampled shearlet transform (NSST) is employed for multi-scale decomposition of the source images, yielding their respective base layer and detail layer images. Guided by the saliency matrix and edge decision matrix, the base layers are then fused using an information entropy weighting method, while the detail layers are fused using a dual-channel spiking cortical model (DCSCM). Finally, the merged base and detail layers are reconstructed to produce the final fused image. According to experimental results, the framework successfully produces fused images with rich information, distinct texture details, and highlights for infrared targets.