The fusion of infrared and visible images via decomposition-based structure transfer and local saliency detection

Abstract

Most of existing image fusion methods neglect the differences between multi-source images and extract characteristics without any discrimination, which negatively impacts the fused result, particularly in infrared (IR) and visible image fusion. This paper proposes a two-scale fusion method using total variation (TV) model with structure transfer and local saliency detection. Firstly, we use rolling guidance filter (RGF) to perform two-scale decomposition on the input images. Secondly, we combine the base layers with TV minimization based on structure transfer. Then we construct the weight map with local saliency detection and guide filter, and combine the detail layers by weight average. At last, the fused result is reconstructed with the fused base layer and the fused detail layer. The fusion result retains more target location and appearance information, which is beneficial to subsequent processing. The comparisons with 7 other methods reveal that our method achieves a superior performance.