An Inpainting Model of Fractal Group Sparse Representation Combined With Residual Denoising Network

Aiming at the blurring artifacts in inpainting, this paper proposes an inpainting model of fractal group sparse representation combined with a residual denoising network. On the one hand, the fractal group sparse representation model can use a unified framework to describe the local smooth and nonlocal self‐similarity features of the image to complete rough inpainting, providing important basic image information for the residual denoising network. On the other hand, regarding the denoising subproblem of fractal group sparsity, the residual denoising network can provide the fractal group sparse representation with detailed information, obtain an adaptive dictionary, optimize repaired details, reduce blurring artifacts, and achieve fine restoration. Based on fractal group sparse representation, this model introduces a residual denoising network for detail optimization, improving image sharpness and reducing artifacts. The experimental results demonstrate that the proposed model outperforms existing methods, including IRJSM, IRGSR, IRCNN, and IRFDnCNN, by achieving the best PSNR and SSIM scores, with average improvements of 2.3700% and 0.4746%, respectively.