Blur Kernel Estimation Using Structure Edge

The blur is usually caused by the relative motion or inaccurate focusing.Blur makes the image with low contrast and detail lost,affects the image perception,brings difficulties to the subsequent processing.Deblurring is a technical that recover sharp image from blurred one.There are many applications in the production and life.Deblurring development in recent ten years is very fast,especially after many studies indicate that the structure edge is important for blur kernel estimation,a number of effective algorithm is proposed.However,these algorithms have some shortcomings in the separation of structure edges and the estimation of blur kernel.In this paper,the blur kernel estimation algorithm and the separation structure edge algorithm are studied.This paper focuses on the following two aspects: on the one hand,designs the new objective function of blur kernel estimation,on the other hand,proposes an image smoothing algorithm for keeping the edge contrast.Specifically:(1)We introduce the concept of blurred structure,solving the imbalance objective function in kernel estimation based on structure;then using sparse prior of blur kernel,designs L0 norm constraint regularization on amplitude domain and gradient domain of blur kernel,solves dense kernel problem.(2)A scale based image scaling operation is proposed,which can preserve the structure while removing the texture.We design and guide the L0 smoothing algorithm,and introduce the guiding map which represents the edge confidence of the structure,and determine the edge which needs to be preserved.After scaling,the map will be used as guidance to the guided L0 smoothing algorithm.Two innovation points are combined to form the scale guided L0 smoothing algorithm,which solves the problem that the high contrast texture is difficult to remove and the edge is blurred after the image smoothing.