Restoration Of Multiple Images With Image Fusion

The research of image restoration can be traced back to the early times of image processing,but it is still challenging because of the unknown course of image degradation and the ill-posed reverse process.Thus image restoration is still an active area of research in computer vision/image processing.In the fundamental theory of image degradation,the process of image restoration can be divided into two stages:Point Spread Function(PSF)estimation and non-blind deconvolution.This article includes innovative progress in both stages.The current methods of PSF estimation can use single image or multi-image.The method proposed in this article will use the multi-image one,which adopts a piece of rapid-exposure noisy image to restore multiple motion-blurred images.By extracting the transparency of locally identical objects in blurred-images and the noisy image, the algorithm in this article can effectively estimate the PSF of the blurred images.Given the PSF,non-blind deconvolution can be adopted to obtain clear images. The non-blind deconvolution is a respectively mature research area.However,studies show that,even if the PSF is known,the state-of-the-art deconvolution algorithm still cannot avoid introducing artifacts in the restored image.Based on the study of image fusion,the innovative practice in this article is to add image infusion into classical deconvolution to deblur the images,and the artifacts produced in the process of deconvolution are inhibited.Experiments show that the final image produced by this method is better than that of the classical deconvolution.Innovations in this article can be summed up as:1.Fully takes the advantage of abundant information of multi-image,and computes the transparency of the same local objects in both burred images and the noisy image.The PSF of blurred images can be estimated using the transparency.This operation allows the input and restoration of multiple blurred images of the same scene.2.Joins image fusion into the deconvolution stage to inhibit the ringing artifacts which are introduced by the deconvolution.The resulted images are clearer than that processed with traditional methods.3.Computes with local objecrs transparency instead of full image's luminance value that reduces the amount of calculation and makes the processing of local blurred images possible.The experiments demonstrate that,methods proposed in this article are effective and can achieve better restoration effect.