Image Inpainting Algorithm Based on KSVD and Improved CDD

In order to solve the problem that the traditional overcomplete dictionary can't repair the image well when the image data is sparsely coded, this paper uses a learning-type overcomplete dictionary called K-Singular Value Decomposition (KSVD) to enhance the adaptive of algorithm. At the same time, for the problem that the visual coherence of the repaired area is poor and the fuzzy area still exists, the curvature-driven diffusion (CDD) technology is designed to further deal with the damaged area. Finally, using the proposed CDD-KSVD algorithm to repair the image in YUV color space. Experimental results show that compared with the image restoration algorithm using KSVD and the image restoration algorithm based on CDD, the CDD-KSVD algorithm proposed in this paper recovers the details of the damaged area better, and the repaired image is more in line with human visual perception, at the same time, we get higher values through objective metrics PSNR and SSIM.