Research On Single Image Super-Resolution Reconstruction Based On Sparse Representation

Instead of changing the existing physical device, image super-resolution technology can obtain the required high resolution images just by employing an appropriate digital signal processing technology. For its’advantages in technical and cost, it has being more and more used in the fields of high-definition digital TV, military remote sensing monitoring, public safety and medical imaging, etc. Compared to multi-frame super-resolution reconstruction technique, single-frame image super-resolution technology only utilizes one low-resolution image of the actual scene to estimate the required high-resolution image of the same scene in image reconstruction. So it can better meet the actual application requirements in some applications. Sparse representation based image super-resolution reconstruction method is a new orientation for single-frame super-resolution image reconstruction technology. Therefore, the research objective of this thesis is single-frame super-resolution image reconstruction technology, especially sparse representation based image super-resolution reconstruction.In this thesis, the algorithmic models and fundamentals of single-frame super-resolution image reconstruction technology is firstly described, the principles, concrete implementation methods and performance characteristics of several classic super-resolution algorithms are analyzed and discussed in detail, and then experimental simulation is conducted for their comparative analysis both subjectively and objectively.Secondly, deep research on sparse representation based super-resolution method is conducted. Research results show that dictionary size and dictionary composition has a great impact on the algorithm’s reconstruction quality and reconstruction efficiency. Moreover, regularization parameter selection method influence reconstruction quality. Experiment and simulation result in a conclusion, that is, with a certain number of dictionary size, to design an algorithm by changing the dictionary structure and regularization parameter may improve the running efficiency of the algorithm as well as ensure the quality of the reconstructed image.Finally, a K-means clustering improved initial cluster center based dictionary constructed method is designed and then applied to reconstruct super-resolution image. Furthermore, in the process of image reconstruction, as for the block effect existed in the edge of the reconstructed image when adopt a sub-dictionary, the analysis and discussion on the choice of clustering sub-dictionary number and the selection of sparse representations coefficient regularization parameter is conducted to reduce blocking artifacts so as to improve the quality of image reconstruction. Experimental results show that, the algorithm in this thesis improves the operation efficiency of the algorithm while ensure the quality of the reconstructed image. Meanwhile, compared with the conventional method, the K-means clustering improved initial cluster center based dictionary constructed method enhances the algorithm’s stability.