Research on image demoiréing method based on multiscale mix structure

With the popularity of smart devices, people often use smart devices to recapture images on multimedia platforms. Moiré patterns generated in the recapture process are an interference texture, and its complexity and unevenness seriously affect the image quality and visual effects. To this end, this paper proposes a multi-dconv parallel enhanced removal block (MPERB) and a multi-dconv parallel enhanced tone mapping block (MPETMB). On this basis, a multiscale mix structure demoiréing network (MuMiDeNet) is proposed to effectively remove moiré patterns while restoring image details and colors. According to the experimental findings on the FHDMI and UHDM datasets, the PSNR and SSIM values outperform those of the current techniques, including MopNet and MDDM, in the quantitative comparison. In the qualitative comparison, MuMiDeNet not only effectively removes moiré patterns, but also enhances the local details and global tone performance of the image. It also performs better than existing methods such as MopNet and MDDM in visual effects and detail restoration, achieving the best results.