An Adaptive Super-resolution Image Reconstruction Method

With the development of Internet technology,image has become an important medium of the modern information system.It plays an important role in our life and scientific research.In particular,the emergence of new electronic products,we put forward higher requirements on the quality of the image,no longer limited to ordinary pixel images.Due to the limitation of industrial technology,the image resolution is reduced due to some unpredictable factors in the process of image generation,transmission and storage.The main performance of the image noise pollution,directly affect the human visual image of the important information acquisition and recognition.In the real world,there are two kinds of common noise,Gauss and Laplace,for the case of single noise,the use of some classical algorithms can get a better reconstruction effect.The regularization algorithm based on L1 norm has a good effect on the Laplace noise,which is a good edge preserving effect,the regularization algorithm based on L2 norm has a good effect on the Gauss noise,which is a good smoothing effect,but also smooth out the details of the edge.In general,the low resolution images generally contain mixed noise,and the regularization algorithm based on L1 norm or L2 norm can not achieve good reconstruction effect,which brings forward higher requirements for the image reconstruction algorithm.Therefore,we study and innovate on the basis of super-resolution image reconstruction.The main work of this paper is as follows:(1)An adaptive data fidelity model is proposed:Considering the low resolution images often contain Gauss and Laplace two kinds of Gauss and Laplace noise,in the process of image reconstruction,the noise distribution ratio will change.Therefore,in this paper,the maximum likelihood estimation method is used to estimate the standard deviation of Gauss and Laplace noise to construct an adaptive membership function.Combined with L1 and L2 norms,an adaptive L1-L2 norm model is designed for the data fidelity of the objective function.(2)An adaptive regularization parameter model is designed:Considering that the image reconstruction is a morbid problem,in order to obtain the only stable solution,in this paper,we adopts the bilateral total variation based on the Li norm form as the regular term.It has the ideal sparse solution,high computational efficiency,it can be a good edge preserving and noise suppression.In particular,the regularization parameter is used to balance the relative contribution of the data error term to the regular term,it is essential for the overall effect of image reconstruction.In the iterative computation process,the noise distribution ratio will change,and the regularization parameter determined by the manual experience can not accurately describe the trade-off between the data fidelity term and the regularization term.Therefore,we present a simple and effective adaptive regularization parameter method.Compared with the traditional manual experience,the experimental results show that the method has certain advantages.