Infrared and Visible Image Fusion Based on Saliency Detection Dynamic Threshold Neural P Systems in Non-Subsampled Shearlet Transform Domain

Infrared and visible image fusion is the process of integrating information from multiple sensor images to obtain a comprehensive understanding. However, existing fusion algorithms face challenges in achieving high-quality fused images. In this article, a novel fusion framework based on saliency detection dynamic threshold neural P (SD-DTNP) systems is proposed. The non-subsampled shearlet transform (NSST) is first introduced to obtain key features at different scales and orientations. To overcome the disadvantage of visual saliency map (VSM)-based algorithms resulting in unnatural fused images, a novel fusion rule of the contrast saliency DTNP (CS-DTNP) system is designed to fuse the low-frequency sub-band. The high-frequency sub-bands are fused by combining the multiscale morphological gradient (MSMG) with the DTNP system, which can avoid the influence of noise and integrate more structural information. Experiments demonstrate that, compared with seven advanced fusion algorithms, the proposed algorithm not only improves the spatial consistency of the final image but also preserves the salient and detailed information.