Infrared and Visible Image Fusion via Double Fusion Branch

Despite the impressive achievement of recent image fusion methods, they overlook a crucial factor that jointly influence the fusion result from the intermediate feature and output side: intermediate fusion result. In this paper, to activate more pixels for better fusion, we propose a novel intermediate fusion branch module (IFBM) to efficiently fusing image features in the global and local range for image fusion. Then we propose a novel double fusion branch image fusion framework called DFBF to achieve efficient image fusion. Specifically, DFBF first use of encoder extracted shallow features and deep features, and then via intermediate fusion branch and main fusion branch jointly guided con-text information fusion. Furthermore, We jointly minimize the main fusion branch loss and intermediate fusion branch loss to better pull the fused image to closer to the ideal fused image and push the fused image to far away from the non-ideal fused image. Experimental evaluations conducted on public datasets RS, demonstrate that our proposed method DFBF, surpasses state-of-the-art methods both qualitatively and quantitatively.