基于Bayesian理论的压缩视频超分辨率重构算法

为了对视频进行更有效的压缩,首先建立起一个从原始图像到压缩视频的成像模型,然后在此模型基础上运用Bayesian估计理论,在最大后验概率准则下表述该问题;最后通过综合使用CCD（cyclic coordinate decent）和SA（successive approximations）等方法从理论上给出了压缩视频超分辨率重构问题的一般解决方法,同时针对成像过程中原始高分辨率图像的降质函数一般为未知的情况,提出了一种基于EM（expectation-maximization）算法的降质函数的确定方法.实验结果表明,该算法不仅在峰值信噪比和重构效果方面对压缩视频有较大提高和明显改善,而且该算法易于扩展,具有广泛的应用范围.

Algorithm of Compressed Video Super-resolution Restoration Based on Bayesian Theory

At first an acquisition of the compressed video model is proposed,then the super-resolution problem is for mulated within the Bayesian framework and the MAP(maximum posterior probability) criterion,finally a universal solution of the problem is presented by integrating the CCD(cyclic coordinate decent) with SA(successive approximations).In order to resolve the problem that the original high-resolution's quality-reduce function is always unknown,a new estimation method is introduced based on the EM(expectation-maximization) algorithm.The results of the experiment demonstrate that the algorithm not only outperforms the traditional ones on the aspects of PSNR and restoration vision effect,but also has the characteristic of easy extension.