Super-Resolution Reconstruction Based On Sparse Representation

In recent years, image processing methods based on sparse representation have quickly become one of the research focuses in the field of image processing.Research on the methods of getting optimal solution and constructing dictionaries is the key to solve the image processing problems based on sparse representation. Image super-resolution reconstruction is an important research branch of image processing, which has been widely used in the fields of the satellite remote sensing, medical imaging, military reconnaissance, video monitoring and so on. It has not only theoretical research value but also practical application significance for us to do research on image super-resolution reconstruction combined with sparse representation.Firstly, this article shows and realizes several kinds of methods of super-resolution reconstruction, which are commonly used. After that, this article focuses on image super-resolution reconstruction based on sparse representation.The whole work for this article includes the following five aspects:1. Do research on three kinds of typical super-resolution reconstruction methods: interpolation methods, the methods based on frame reconstruction and the methods based on learning. And implement several super-resolution reconstruction methods on MatLab 7.11.0 platform.2. Study on sparse representation model, the methods of getting optimal solution, the methods constructing dictionaries, sparse dictionary model and morphological component analysis (MCA) model.3. Propose and realize two super-resolution reconstruction methods based on sparse decomposition. Do morphological component analysis to the image, and get the cartoon component and texture component of the image. Then use Self-snake model or C&E model to reconstruct cartoon component, and use the bicubic method to reconstruct texture component, based on the fact that the cartoon and texture have different characteristics.Add the cartoon component and texture component after the reconstruction, the reconstructed image can be got. The experiment results show the combination of morphological component analysis method and interpolation methods can offer us better reconstructed image.4. Combining image super-resolution reconstruction methods based on sparse representation with sparse dictionary model, the image super-resolution reconstruction model based on multiple sparsity is proposed. And use it to reconstruct image.The experiment results show the new model can not only improve the efficiency of training dictionary but also recover more high-frequency details has been lost.5. After subjective evaluation, do the objective evaluation with the mean square error (MSE), peak signal to noise ratio (PSNR), gray offset curve and structural similarity (SSIM) to the reconstruction result. 属性不符...