Dual Discriminative Relative-of-Gaussians Prior for Blind Deblurring
Yufan Chen, Xinru Chen, Zihan Cheng, Wenze Shao · 2023
This paper proposes a novel dual discriminative image prior for blind deblurring. The desired discriminativeness between salient edges and faint textures as well as a sharp and blurred image has been both elegantly formulated via use of relative-of-Gaussians. Utilizing the prior in the maximum-a-posterior estimator, blind deblurring is then transformed into a simple yet intuitive image filtering task via the well-known half-quadratic regularization (HQR). Experimental results on both synthetic and realistic blurred images well demonstrate the practicality of the proposed prior and also its comparable or superior performance to existing state-of-the-art ones.