Polarization image fusion method based on polarization saliency with generator adversarial network

Abstract

Aiming at the problem that the polarization image fusion process does not make good use of the polarization imaging characteristics, we propose a polarization image fusion method based on polarization saliency. On the one hand, we propose a new definition of physical property characterization based on polarization properties:polarization saliency. We construct it from three perspectives: image difference, position, and contrast. On the other hand, we propose a polarization saliency based fusion network (PSGGAN). The information decision block is constructed to enhance the polarization salient information and texture detail information of the source image. The loss function based on polarization saliency is designed to regulate the pixel distribution constraints and the degree of polarization information preservation. Qualitative and quantitative experiments demonstrate the superiority of our PSGGAN over state-of-the-art methods in terms of visual effects and quantitative metrics.