A multi-scale infrared polarization image fusion method based on polarization-forming

In traditional infrared polarization image fusion, the degree of linear polarization (DoLP) image is frequently fused with the intensity image ($S_0$). However, it is easily interfered by background noise, which degrades the quality of the fused image. To address this problem, a multi-scale infrared polarization image fusion method based on polarization-forming (PF) is proposed. Aiming at the sensitivity of the DoLP image to background noise, the approach is proposed to strip the information of the input image. The decomposition of the DoLP image is guided by controlling the inputs to the Stokes model. Subsequently, the ratio of local entropy (RoLEN) is employed to obtain an objective polarization vector image ($OPV$) for noise suppression. For the blurred detail, a fusion framework based on multi-scale hybrid window filtering (MSHWF) is constructed to fuse $OPV$ with $S_0$. The framework adopts an improved filtered differential structure to separate edge and weak texture information. Then fusion strategies based on adaptive fractional saliency detection and adaptive energy are applied to highlight the target and its detailed information, respectively. Finally, the extracted sub-layer signatures are fused to obtain enhanced images with clear objectives and rich details. Qualitative and quantitative experimental results demonstrate the superiority of our method over other infrared polarization image fusion algorithms in several metrics such as information entropy, average gradient, and spatial frequency.