Fast Image Super-resolution Reconstruction Based On Sparse Representation And Regression

Image super-resolution reconstruction technology is to improve the resolution of the original image by means of signal processing without changing the imaging device,thus enhancing the visual effect of the image for later analysis and processing.As the technology in many areas has a wide range of applications,it has become a hot issue.At present,the mainstream super-resolution reconstruction method is based on the learning method,in which the algorithm based on sparse representation or support vector machine regression has greatly improved the quality of image reconstruction.Yet,these approaches are hard to apply in practice because they are either too slow or demand tedious parameter tweaks.In view of the above problems,we regard the speed and quality of image reconstruction as the research content and propose the method of combining sparse representation and regression analysis to better solve the image super resolution problem in this paper.In order to improve the speed of image reconstruction,we merge the idea of neighborhood embedding and sparse representation in this paper.Firstly,the atoms in the dictionary are used as the neighborhood space,and then the image is quickly reconstructed by neighborhood regression projection.Experimental results show that the sparse representation and regression method are effective and the running time of reconstructed images is reduced.Based on the above work,we propose a fast reconstruction algorithm based on sparse representation and linear regression to further improve the quality of image reconstruction.Firstly,a dictionary is trained by K-SVD algorithm based on training samples.Then the entire data set can be divided into a number of sub-spaces according to the atoms in the dictionary.Moreover,the mapping from low to high-resolution images can be obtained independently for each sub-space.Finally,the high resolution image is quickly and accurately reconstructed by selecting the appropriate projection matrix.Experimental results demonstrate that both the image reconstruction quality and the speed of our proposed algorithm perform better than other popularly used methods.