Image Restoration Technique Based On Compressed Sensing

The image restoration has been a hot research technique, it plays an important role in many disciplines and fields,such as computer vision, medicine,space technology,and digital image and signal processing.Image denoising and deblurring are the important research direction in the image restoration, it requires the degraded image not only can remove noise and fuzzy, but also as far as possible to retain the details of the image texture and edge, but they often cannot be simultaneously satisfied. The compressed sensing technology is applied to image restoration, and the existing algorithms are improved and innovated to achieve the effect of restoration.The research task is divided into two main points.Generally in the image sparse representation,the global dictionary can not expressed the part of details very well.In order to improve this situation, the algorithm in this paper is that the image is divided into three parts,namely, the smooth, texture and edge class according to the characteristic of the image firstly. The thought of the decomposition is that the training samples are divided into flat and smooth category according to the size of the variance,and the flat category is divided into texture and edge classes using the method of principal component analysis.Using the K-SVD dictionary training learning methods to train the three parts are be decomposited,In order to obtain an adaptive redundant dictionary, and sparse representation to achieve the purpose of denoising.The final image and the results show that the improved algorithm based on classification of the image and adaptive redundant dictionary to denoising compared to the other two algorithms can improve the effect of denoising,and in some of the details of texture and edge has been improved,the visual effect to people is also clear.The general algorithms of image deblurring require that the point spread function is known, but this paper uses the method of K-SVD dictionary learning and image classification to study the blind image.Firstly set an initial point spread function,and then combined with the dictionary learning to get the point spread function,finally the blurred image by deconvolution to achieve the purpose of deblurring. Based on the research of image processing about the motion blur and Gauss blur,finding that the algorithm in this paper has better effect in image deblurring.