Visible Light Polarization Image Fusion based on Dense Connection Generative Adversarial Network

Abstract

Polarized images have rich high-frequency information. The texture, contours and edges of objects in the image are very obvious, while the intensity images contains the main energy of the image and the background field is rich in information. Therefore, the polarization image is combined with Light intensity image fusion is of great significance. This article uses deep learning methods, based on the training method of generating adversarial networks, using densely connected generator networks, using SSIM loss and gradient loss functions, and experiments have shown that the ideal fusion effect can be achieved.