Super-Resolution Image Reconstruction Based On Multi-Component Dictionary Sparse Representation

The main research problem of the super-resolution image reconstruction is given a picture of a low resolution image get high resolution images of the same scene by certain method. The sparse representation for image reconstruction has the advantages of the required quantity is less and accurate image reconstruction quality of the reconstructed images. The key problems of sparse representation for super resolution image: the fast training sparse dictionary and the super resolution image reconstruction precision.In this paper, the research of the super resolution image reconstruction based on sparse representation, the main work lies in the following aspects:Firstly, the structure of complete dictionary of super-resolution image reconstruction. Because of the reconstruction process of the image is that image is reconstructed by iteration of the image block. This requires the high and low resolution image blocks to achieve a certain degree of registration requirements. This paper puts forward the idea of the introduction of double resolution dictionary based on traditional dictionary training. Through our dictionary training algorithm to optimization reconstruction image blocks' sparse coefficient on the dictionary, the high and low resolution image block registration is higher. In this paper, the KSVD method is used for the training of the dictionary. By combining a dictionary structure and K-SVD dictionary training methods, to obtain a more complete dictionary generation method.Secondly, the richness of dictionary is often the key factor to determine the quality of the reconstructed image, so this paper proposed a multi-component(smooth, texture, edge) adaptive dictionary selection scheme, improved the quality of the dictionary and the quality of the image reconstruction.Finally, the reconstruction of super resolution image on low resolution image to high resolution image. Because of the question that algorithm complexity of super-resolution image reconstruction is great, leading to reconstruction time is too long. This paper puts forward a kind of optimization model based on norm. Getting high resolution image based on the regularization collaborative representation for low resolution image in the multi-component adaptive double dictionary. The complexity is reduced greatly and the reconstruction speed is improved greatly without loss of local characteristics.The reconstruction results show that this kind of dictionary construction method to reconstruct the image is more expressive, for traditional evaluation indices PSNR and RMSE has better performance. Whether it is in the quality of reconstruction, or in the reconstruction speed than traditional algorithm has significantly improved. The validity of the algorithm is verified.