An infrared and visible image fusion method with edge-preserving and high-speed

Abstract

Infrared Fused images generated by infrared and visible image fusion algorithms often suffer from problems such as blurred edges and overlapping of targets and backgrounds, mainly because most existing fusion methods process images from a global perspective without special consideration for edge preservation. Therefore, to obtain fused images with clear edges, gradient information should be emphasized during the fusion process. Meanwhile, to realize real-time display of fused images and support high-speed processing for downstream tasks, the computational complexity of fusion algorithms urgently needs to be reduced. Based on the EgeFusion algorithm, this paper proposes an edge preserving high-speed image fusion method. The proposed method uses weighted guided filtering to achieve two-scale image decomposition, dividing the original image into a base layer and a detail layer. The slope coefficients of the guided filter are directly utilized to construct the fusion weights for the detail layer, which avoids the traversal computation of sub-window variance, significantly reduces computational cost while retaining edge gradient information. For the base layer, the same visual saliency weighting strategy as EgeFusion is adopted to ensure the global structure and target saliency. Experimental results on the RoadScene dataset show that the proposed algorithm outperforms EgeFusion in terms of metrics such as mutual information, peak signal-to-noise ratio, Qabf, and structural similarity. The fused images have clear edges and prominent targets, and the running speed is improved to 6 times that of the original algorithm. The proposed method can meet the real-time processing requirements of engineering scenarios while ensuring fusion quality