Research On Image Super-resolution Reconstruction Based On Sparse Representation

In recent years, the sparse representation theory has developed rapidly in the fieldof image processing, which has been successfully applied to the image super-resolutionreconstruction. On the basis, this paper puts forward two very effective methods: Theone is image super-resolution reconstruction algorithm based on MCA(MorphologicalComponent Analysis) and dictionary learning，the other is image super-resolutionreconstruction algorithm based on self-similarity and double sparsity dictionary. Bothmethods can improve image quality better.The image super-resolution reconstruction algorithm based on MCA and dictionarylearning introduces MCA image decomposition theory in image reconstruction process,which can get the high frequency texture information of the image better. In thismethod，an image is viewed as a linear combination of the structure part and the texturepart. Due to the different properties between the structure part and the texture part,different image super-resolution reconstruction methods are used to reconstruct themrespectively. And then the expected high-resolution image is obtained by combining theboth.The image super-resolution reconstruction algorithm based on self-similarity anddouble sparsity dictionary can achieve real-time, fast reconstruction without externalnatural image library. In this method, High-and low-resolution images are obtainedfrom the image pyramid which is formed by the self-similarity of the low-resolutionimages between different scales and the same scale. Then the high-and low-resolutiondictionary pairs are achieved by the method of double sparsity dictionary learning.Finally，the expected high-resolution image is got by reconstructing the low-resolutionimage.Experimental results show that the two methods proposed in this paper are able toobtain better reconstruction results. The image super-resolution reconstruction algorithmbased on self-similarity and double sparsity dictionary can obtain a high-quality image,which preserves the edge and texture information well. While the imagesuper-resolution reconstruction algorithm based on self-similarity and double sparsitydictionary can not only achieve real-time, fast reconstruction of an image, but alsoguarantee the reconstructed image quality.