Image Motion Deblurring

Image motion deblurring is a research hot spot in image restoration, and applied abroad in industry, military affairs, medical treatment and daily life. In recent years, lots of algorithms about image motion deblurring have been proposed, however, most of them have limitations such as heavy computation, slow speed, small motion kernel assumption, multi-images, additional hardware, or severe artifacts such as ringing, and all of them make it far from practical use. In this paper, we introduce a fast, robust blind deconvolution based single image motion deblurring algorithm. We use shock filter to predict strong edges of latent image, and with this we can constrain image gradients and motion kernel with simple and easily optimized Gaussian prior, and estimate motion kernel in frequency domain fast and robustly. After that, we use a heavy-tailed distribution of natural images based sparse non-blind deconvolution algorithm to restore the latent image, which can restore the salient edges and texture as well as suppresses ringing and noise. In the motion kernel estimation part, we use conjugate gradient method to optimize the energy function, and use well conditioned image first and second derivatives with small condition number only and exclude pixel value with large condition number to make the algorithm converge fast. Besides, we introduce a hysteresis thresholding to suppress the noise of motion kernel significantly, and use a re-center method to improve the robustness of large motion kernel estimation. The experimental results show that our algorithm can restore high quality latent images with clear edge and texture, little ringing or noise from a single motion blurred image fast and robustly.