Research On Image Restoration Based On Saliency Image Edge Joint Sparse Representation

Images may exist blurring due to moving cameras and mobile device,the losing details and image features bring certain difficulties for image preprocessing.The existing image acquisition device has no obvious effect on the large scale motion deburring.In recent years,image deburring has been became a hot issue in the field of image processing and computer vision.In order to better resist image noise and describe complex textures,the paper mainly study the image restoration based on significant edge joint sparse representation.The main work and innovation in the paper are listed as follows,1.A significant edge extraction algorithm is proposed to preserve texture details.Traditional edge extraction methods using two-valued function to suppress image noise,which may damage the image structure and kernel structure at the same time.What's more,ignoring image details may lead to the adverse effect in the restored images.An adaptive edge extraction operator is proposed in this paper based on data constraint and gradient constraint.In which data constraint to suppress image noise and gradient constraint to texture enhancement.Experimental results show that the image noise are obvious reduced and the more image details are extracted based on the proposed method.2.A gradient constraint method are developed for preserving fuzzy kernel structure.Although the Traditional threshold methods can reduce the image noise,the kernel estimation results may cause adverse effect.In this paper,we apply a gradient constraint to protect the structure features by counting the number of pixels which the gradient values are not zero.Experimental results show that our method can filter the outlier points and mitigate the adverse effect effectively.3.A non-blind deconvolution method is proposed based on sparse representation.Traditional non-blind deconvolution methods are greatly affected by the image noise,and the restored images may exist boundary effects.We propose an image reconstruction method based on sparse representation.We achieve a better adaptability by training dictionary.Experimental results show that our method can restrain the boundary effects and improve the image restoration on texture details.