Infrared and visible image fusion algorithms based on generative adversarial networks

Abstract

Visible images have good contour and texture information, while infrared images have the advantage of working in all weather. Therefore, for the detection and analysis of targets in low illumination at night, the information from visible and infrared images can be fused to improve the detection accuracy and anti-interference capability of detection systems for nighttime targets. In this paper, we propose a generative adversarial network-based fusion algorithm for IR and visible images, which can effectively extract the feature information of IR and visible images by adversarial training of two discriminators and generators, improve the feature extraction ability and the quality of fused images by introducing attention mechanism and structural similarity loss function, and enhance the stability of network training by TTUR. The experimental results show that the algorithm in this paper outperforms other typical algorithms in both subjective and objective evaluations.