The Single Image Super Resolution Based On Sparse Representation And Adaptively Dictionary Learning

Super resolution image reconstruction method offers the promise of overcoming some of inherent resolution limitations of imaging sensors, allowing better utilization of the growing capability of high resolution displays. Such super resolution images may also prove to be essential in medical and satellite imaging where diagnosis or analysis from low-quality images can be extremely difficult. Recently traditional methods are based on the super resolution reconstruction of the same scene in multi-frame images, these images exit subpixel shift. This class of methods is fusing the low resolution images to build a super resolution image. However, without high frequency, the performance of these reconstruction-based super resolution algorithms degrades rapidly if the magnification factor is large.This paper addresses the problem of generating a super resolution (SR) image from a single low resolution input image. We approach this problem from the perspective of compressed sensing:image patches can be well represented as a sparse linear combination of elements from an appropriately chosen over-complete dictionary. Inspired by this observation, firstly we trained two image dictionaries of low and high image space characters, seeking a sparse representation for each patch of the low resolution input in the low image dictionary, and then use the coefficients of this representation and high image dictionary to generate a high resolution output.To enforce the similarity of sparse representations between the low resolution and high resolution image patch pairs, we jointly training two dictionaries for the low and high resolution image patches. Simultaneously adaptive dictionary selection is applied to strength the time efficiency. Compared to previous approaches, our training sets are not necessary to be the same class, which overcomes the reality of the lack of multi-frame low resolution images that can not be used to rebuild high resolution image. The experimental results show that our approach has greatly improved the results in both subjective and objective measure.