Blind Image Restoration Using Divisional Regularization and Wavelet Technique

Abstract

Regularization method has been widely used in blind image restoration. Most regularization operators, however, are applied uniformly without considering difference of edge regions, which results in an unsolved trade-off conflict between smooth and edge regions. In this paper, we apply suitable regularization operators to smooth regions and edge regions respectively according to their characteristics instead of a global and constant one, and further employ the wavelet technique to control the noise amplification in order to improve the quality of divisional regularization. Experiment results show that our proposed Divisional Regularization and Wavelet Technique (DRWT) can deblur effectively without ringing effect in the restored images, thus improves edge restoration and reduces noise amplification.

Image Restoration and Object Removal Using Prioritized Adaptive Patch-Based Inpainting in a Wavelet Domain

“…The authors in [28] proposed CWT processing while the method proposed in [29] works in wavelet transform with the fast marching method (FMM). The approaches proposed in [30][31][32] used wavelet transform techniques for reproducing edge details with various techniques like total variation (TV), divisional regularization, the fast optimization transfer algorithm, etc. mostly for processing and restoring noisy images.…”

“…The method proposed in [33] uses structure propagation by level lines and Bezier curve approximations to reproduce the edges, but it only works for small regions. Most of the algorithms proposed in [27][28][29][30][31][32] reproduce edges and simple textures better for small damaged regions, but they lead to the existence of artifacts in large regions and big object removal.…”