A dual-weighted polarization image fusion method based on quality assessment and attention mechanisms

Abstract

This paper proposes a dual-weighted polarization image fusion method based on quality assessment and attention mechanisms to fuse the intensity image (S0) and the degree of linear polarization (DoLP). S0 has high contrast and clear details, and DoLP has an outstanding ability to characterize polarization properties, so the fusion can achieve an effective complementation of superior information. We decompose S0 and DoLP into base layers and detail layers. In the base layers, we build a quality assessment unit combining information entropy, no-reference image quality assessment, and local energy to ensure the fused image has high contrast and clear and natural visual perception; in the detail layer, we first extract depth features using the pre-trained VGG19, then construct an attention enhancement unit combining space and channels, and finally effectively improve the preservation of detail information and edge contours in the fused image. The proposed method is able to perceive and retain polarization image features sufficiently to obtain desirable fusion results. Comparing nine typical fusion methods on two publicly available and own polarization datasets, experimental results show that the proposed method outperforms other comparative algorithms in both qualitative comparison and quantitative analysis.