Research On Adaptive Super-Resolution Image Reconstruction Algorithm Based On Image Blocks

Super-resolution image reconstruction is a technique to reconstruct a high-resolution image based on a single or a series of low-resolution images of the same scene. Now it is widely used in remote sensing, military reconnaissance, pattern recognition, medical image processing and other fields. In recent years, as the theory of compressed sensing was introduced to the image processing and other related areas, the sparse theory is also applied to super-resolution image reconstruction. And the better results have been achieved than traditional methods of image reconstruction.Sparse representation is a super-resolution image reconstruction method based on learning method. It not only can take advantage of features of an image itself, but also can make efficiently use of the priori information of image by training dictionary. Therefore, it is widely used in image super-resolution reconstruction of single image. This thesis focuses on dictionary learning method based on sparse representation of single image super-resolution reconstruction.For dictionary training and reconstruction, a two-layer K-SVD dictionary learning and reconstruction method for super-resolution image reconstruction is adoped to solve the problem of limitations of size of dictionary and the input low-resolution image details. Firstly, the first level dictionary is trained based on the training library. Then, for a low-resolution image, its reconstructed image is obtained based on the first level dictionary and it is used as input image of second level dictionary. As the input image of the second level dictionary have more detail information, the reconstruction of image can recover more detail information and have a better result. In addition, the sparse representation is replaced by collaborative representation for image reconstruction to avoid solving sparse coefficients in sparse coding in this thesis. Thus, the complexity of the computation is reduced and the image construction time is decreased. The frame of two-layer dictionary learning with iterative back projection(IBP) is optimized to improve the quality of image reconstruction.For sparse decomposition, an adaptive of image reconstruction algorithm based on image blocks is proposed in this thesis. The low rank characteristics of similar image blocks are used and a super-resolution image is reconstructed based on iterative algorithm of singular value decomposition. In order to solve the problem of complex computation and slow speed of composition of the algorithm, an adaptive of image reconstruction algorithm is proposed. The image reconstruction can choose the algorithm adaptively based on image block feature of variance to improve the speed of image super-resolution reconstruction.The experimental results show that the proposed algorithm can greatly decrease the image reconstruction time and barely reduce the reconstruction precision.