联合信息增强与生成对抗网络的红外与可见光图像融合

Infrared and visible image fusion of joint information enhancement and generative adversarial networks

为克服红外与可见光图像的融合结果图边缘模糊、细节信息表征不足及亮度较低的问题,提出了联合信息增强与生成对抗网络(Generative Adversarial Network,GAN)的红外与可见光图像融合方法。基于自适应直方图划分和亮度校正算法被用来增强红外图像的显著对比度。多尺度Retinex算法(MSR)被用来增强可见光图像的纹理细节,同时调整可见光图像的亮度。增强后的红外与可见光图像被输入GAN进行训练,训练好的模型可以完成端到端图像融合任务。仿真实验结果表明所提算法在EN(信息熵)和SD(标准差)指标上具有明显优势。

In order to overcome the problems of blurred edges,insufficient detail information and low brightness in the fused image of infrared and visible images,a combined information enhancement and generative adversarial network(GAN)approach for infrared and visible image fusion is proposed.Algorithm based on adaptive histogram division and brightness correction is used to enhance the significant contrast of infrared images.Multi-Scale Retinex algorithm(MSR)is used to enhance the texture details of the visible image while adjusting the brightness of the visible image.The enhanced infrared and visible images are fed into a GAN for training,and the trained model can accomplish the end-to-end image fusion task.The simulation experimental results show that the proposed algorithm has advantages in both subjective visual quality and objective evaluation metrics.