A competent image denoising method based on structural information extraction

Miaowen Shi, Linwei Fan, Xuemei Li, Caiming Zhang

Abstract: The critical problem of image denoising is removing noise while remaining the complex structures of the restored image as much as possible. Therefore, the reconstruction of image structures influences the quality of denoised images. In this paper, we first develop a structure extraction model that detects image structure efficiently. Then the model is applied to stack the similar patch group matrix. Different from other patch grouping methods, this model focuses on the image structure similarity among patches. After this set, we introduce a novel denoising model that incorporates low-rank and kernel Wiener filter priors based on the structure extraction model. The new model takes full advantage of the corresponding patches and remains the fine details as much as possible. Furthermore, the proposed method can reduce the artifacts which are inevitable in most denoising methods. Finally, the optimization problem is solved by alternating direction method multipliers. Experimental results demonstrate the out-performance of the proposed method concerning numerical and visual measurements.