Blind Image Deblurring with Extreme Gradient and Dark Channel Priors

This paper presents a novel blind image deblurring algorithm based on extreme gradient and dark channel priors. Traditional methods relying solely on local maximum or minimum gradient priors to estimate the latent image often suffer from ringing artifacts or loss of high frequency information in low gradient areas. To solve those problems, we combine local minimum and maximum gradient prior information to better constrain the solution space. Experimental results show that the proposed algorithm achieves better restoration performance with detail preservation, noise suppression, and robustness on blurry images including face, low light scene, and text. In addition, our algorithm outperforms other methods on Levin and Köhler datasets, with significant improvement in PSNR and SSIM.