Joint target geometry and polarization properties for polarization image fusion

Traditional polarization image fusion focuses on mining information from the source image and realizes polarization image fusion by finding the feature balance point between two source images. However, this method is strictly limited to high quality and sufficiently informative source polarization images. In addition, if the fusion rules are formulated in terms of the dimensions of the source image information, it is difficult to achieve accurate and efficient fusion results. Based on this, this paper designs a new polarization image fusion model that joins the target geometry and material polarization characteristics from the starting polarization characteristics of different materials (GM-PFNet). In this implementation, a weak reference polarization image quality assessment method (WR-PIQA) for joint target geometry and material polarization characteristics is designed by exploring the imaging laws of scene target geometry and material polarization characteristics. This method utilizes the polarization parameters of target surface roughness, specular reflection, and diffuse reflection, distinguishing it from traditional feature extraction methods. It employs image quality assessment weighting to explore new information from the source images. Since the model belongs to a data-driven self-evolving training model, it is able to utilize the obtained intermediate fusion results to further co-supervise the fused images during the training process. In this way, our fusion results can benefit both from learning from the original input image and from the intermediate output of the network itself. A comparison with state-of-the-art methods on both self-constructed and publicly available datasets reveals that our GM-PFNet model achieves superior performance in both qualitative and quantitative experiments.