Research On Image Super-resolution Reconstruction Based On Sparse Representation

Single image super-resolution reconstruction researches on how to reconstruct a clear high-resolution image from a low-resolution image. With the rapid development of the sparse representation theory, it has been successfully applied to image super-resolution and has caught the attention of a mass of researchers. On this basis, this paper puts forward two effective modified image super-resolution methods based on sparse representation. One is image super-resolution reconstruction based on image decomposition and sparse representation. Another is image super-resolution reconstruction based on image patches classification and sparse representation.The super-resolution based on image decomposition and sparse representation uses the decomposition method based on Mumford_Shah and G field to decompose the image into structure part and texture part. And then reconstruct the texture part with super-resolution reconstruction method based on sparse representation, which can recover the high frequency part more effectively. But the structure part is reconstructed with bicubic interpolation, which can compute quickly and retain the structure feature.The super-resolution based on image patches classification and sparse representation classifies the image patches into smooth patches and patches with texture, on the basis of the features of GLCM(Gray-level Co-occurrence Matrix). And then it reconstruct the low-resolution image with these two different types of dictionary couples. The reconstruction result is improved obviously because the dictionary couple belong to different class can represent the features of the images in their class accurately.Experimental results show that the two methods proposed in this paper are able to obtain better reconstruction results. The reconstruction images can preserve the rough sketch and texture well, and the edges of the images are clearer and smoother.