Multi-autoencoder with Perceptual Loss-Based Network for Infrared and Visible Image Fusion

Infrared image has the obvious target while visible image has the abundant details. Fusing the infrared image and the visible image aims to reserve both the important information of these two. There is still space for the development in this field. In this paper, we propose a novel multi-autoencoder with perceptual loss based and WMC enhancement fusion network. Firstly, we design a preprocessing module in the framework to enhance the visible images from the source images. Then, to preserve more details and textures, we extract the higher feature map and use the perceptual loss to circulate the Euclidean distance of each feature map between the fused image and the infrared image. To highlight the target of infrared image, we use a Sobel-based gradient loss function to train the proposed network. At last, we use WMC to enhance the outputs from the proposed network. Extensive experiments on both public TNO and RoadScene datasets are conducted to test the performances of the proposed method. The experimental results demonstrate that the proposed method outperforms some state-of-the-art methods both qualitatively and quantitatively.