Image Super-Resolution Via Compressed Sensing

Image super-resolution is a technique that uses a low resolution image to a high resolution one.With the development of science and technique,digital images of higher resolution and definition are required.Although hardware is an important way to improve the resolution,it results in much higher cost.Under some practical circumstances,the accuracy is inaccessible for resolution limitation even the most advanced hardware is used in the system.In addition,for some existing images,higher resolution is required for some special purpose.of the images collected several years ago,due to the reasons for the application,therefore,image super-resolution arises for these reason.After the review of the theory of compressed sensing,several construction algorithms and methods of generating dictionaries are studied.Reconstruction performance is compared in terms of these algorithms and corresponding dictionaries.Especially,we focus on the super-resolution method based on the joint-training dictionary and L₁ algorithm,and conduct experimental to validate our results.In order for comparison,this paper uses two methods to reconstruct the same picture:one is the method we have mentioned in the above,and the other is the traditional interpolation method.Experimental results show that the method studied in this paper based on the joint training dictionary and L₁ algorithm provides a clearer image,and its peak signal to noise ratio?PSNR?is 3dB higher than that of the traditional interpolation method.