Research And Implementation Of Image Deblurring Based On Hyper-laplacian Priors

Image deblurring has been a tough and attractive topic, with great value andsocial significance. However, due to the circumstances of the actual degraded imagesis complex and diverse, such as with an unknown blur, existing noise interference,which usually results in an unsatisfactory recovery. Various process models andalgorithms have been proposed, but they have some certain assumption in general andhave limited applicability. This paper presents a new way to process a blurred image,in which we identify the blurred image as either Motion blur or Gaussian blur firstly,secondly estimate the blur specially, and finally restore the degraded image by using adeconvolution algorithm.Blur type has its own characteristic, blurred image after some processing hassome features to the type of blur, especially for Motion blur. In this paper, the keydistinction process includes the extraction of image gradient power spectrum, aButterworth filter, image binarization and so on, and we use the Radon transformationto identify the blur type. The parameter estimation of blur is difficult, especially thereis noise interference in degraded images, and this paper discusses the way ofestimation to PSF specially. It uses the Radon transformation to estimate the Motionblur, and use blind deconvolution algorithm to estimate the Gaussian blur directly.Experiments show that the way to distinguish the blur type and estimate the blur inthis paper has achieved good results, and it has applicability and inspiration to someextent.The common image deblurring algorithms include Wiener filtering, L-Ralgorithm, constrained least squares filtering, blind deconvolution algorithm, we useFast Image Deconvolution using Hyper-Laplacian Priors, and it belongs to calculus ofvariations, while in which the Hyper-Laplacian Priors model is more close to thegradient distribution of natural images, so it can achieve a more accurate result, whilecomputationally efficient, and better performance than the traditional algorithms. Thecombination of this algorithm to the overall process in the paper achieves efficientand ideal recovery. In this paper, the experiments show that the overall process of deblurring canachieve a recovery whit high quality and low time-consuming. The exploration andresearch in this paper is practical and constructive in image deblurring field.