Fraction-Order Total Variation Image Blind Restoration Based on Self-Similarity Features

Abstract

To improve the artifacts of the restoration results restored by existing blind restoration method, an effective image blind restoration method using self-similarity as prior information is proposed for restoring the blurry images. Firstly, the fraction-order model is achieved by extending integer-order total variation, which is prone to reduce artifacts. Motivated by the fact that the introduction of prior information is beneficial to improve the restoration results, we found that natural images usually exhibit some texture features. Self-similarity is a popular texture features and well-defined in the statistics. Therefore, this texture feature is introduced as prior information for the restoration model and further improving the restoration results. Finally, the cost function is generated and solved by semi-quadratic regularization. Experiments on various natural images showed that the proposed method can improve the performance relative to other image blind restoration algorithms in terms of both subjective vision and objective evaluation. The subjective analysis revealed that the proposed algorithm resulted in improved translation and improved artifact appearance. The objective evaluation showed that the proposed algorithm showed the best evaluation values, including Structural Similarity and Peak Signal-to-noise ratio. The restoration results of various images reveal that the proposed method is practical and effective in image restoration. INDEX TERMS Image blind restoration, texture features, fraction-order total variation, prior information.

Fusion network for blur discrimination

“…Blur discrimination is great benefit to the subsequent image processing, including depth estimation, image quality assessment, information retrieval, image restoration, and others. [1][2][3][4] As a result, blur discrimination has already become an important problem in image and video processing systems.…”