Research On Super-Resolution Technology For The Compressed Videos

Recent years, with the increasing usage of the consumer electronic products with photographing function, the wide-spread of high-definition and full-high-definition television, and the establishment of video-surveillance network, the users’ requirements of high-resolution video have risen accordingly. Meanwhile, to achieve better visual experience for people, the modern video displayers with even higher resolution have been designed as a consequence of the development of science and technology. To cope with the inadequacy of video resolution (always been compressed), the super-resolution technology which applies signal processing method, boosts video resolution effectively and satisfies the people’s increasing pursuits of high-definition videos. It can be widely applied in the communities of digital television, internet video, security system, etc.The research background as well as significance of the compressed video super-resolution technology are introduced in this paper. Its recent studying status is summarized, and its main challenge is also discussed. Several important techniques involved in super-resolution reconstruction are listed and described.In this article, an analysis is made for the traditional video super-resolution parameterized model whose main drawback is the ignorance of the non-public regions between neighbor frames caused by the occlusions and overflowed boundary regions. With a parameterized factor, common note matrix, which is applied to distinguish the non-public contents and eliminate the corresponding incorrect reference information, a modified video super-resolution parameterized model is suggested. The calculation of common note matrix bases on the analysis of motion relationship. With the instruction of the modified model, the optimization problem for video super-resolution reconstruction is established based on the Maximum a Posterior (MAP) criterion. It can be solved via Iterative Reweighted Least Square (IRLS) method. A simplified realization for the large dimensional matrix calculation is also introduced. Experiment proves that the modified model based MAP video super-resolution reconstruction method achieves wonderful results, especially around the object edges with occlusion and the image boundary area. The restoration errors caused by incorrect reference information are also avoided effectively.For a real-world video, the modeled parameters are all unknown except for the source sequence itself. Based on the modified model, a mixed MAP video super-resolution algorithm is proposed. The approach integrates the joint estimations of warping matrix, blur kernel, and the super-resolved image into a mixed MAP framework after obtaining the approximation of common note matrix. The precise estimations of unknown elements are realized via alternatingly optimizations iteratively. To speed up the convergence of the optimized solution achieved by IRLS, an improved IRLS iterative weights updating mechanism is suggested which bases on the characteristic of a digital image. A data fidelity constraint function is also designed for better reconstruction result. The experiments demonstrate that the proposed mixed MAP video super-resolution method achieves precise estimations of the unknown modeled factors. It handles test sequences with different content features and diverse scales, and it is also able to cope with different upsampling factors.Given the quantization noise in the compressed videos, the paper suggests a joint estimation algorithmic framework which alternatively removes the quantization noise and does the super-resolution reconstruction. The experiments prove that the proposed compressed video super-resolution method eliminates the quantization noise effectively while boosting the sequences’ effective resolution.