基于DCGAN的可见光与红外图像融合算法研究

he existing visible and infrared image fusion algorithms require the registered visible-infrared image pair in the fusion stage. In many scenarios, this requirement is very strict. In this paper, a DCGAN-based visible and infrared image fusion algorithm is proposed. Through the adversarial training of the generation network and the discriminant network, and the constraints of the specially designed texture loss function, the generation network can directly generate fusion images from infrared images. Experiments were conducted on the FLIR data set. The experimental results show that the fusion images obtained using this algorithm can well combine the color characteristics of visible images and the texture characteristics of infrared images, and the speed of generating fusion images is very fast, which can meet some real-time requirements application.