Abstract:

Compressed video Super-Resolution（SR） technique estimates High-Resolution（HR） images from a sequence of Low-Resolution（LR） observations, it has been a great focus for video SR. Based on the theory of regularization and projection to a convex set, a novel SR algorithm is developed and analyzed using the quantization information from the compressed bitstream. The regularized cost function using the temporal and spatial prior information is proposed. The iterative gradient descent algorithm is utilized to reconstruct the HR image. The reconstructed HR image projects to a convex set in the DCT domain. Experimental results demonstrate that the proposed algorithm has an improvement in terms of both objective and subjective quality, and it is applicable for compressed images.

摘要： 压缩视频超分辨率（SR）技术利用压缩后的低分辨率（LR）图像序列来重建高分辨率（HR）图像的技术，是当前视频超分辨率技术研究的热点。在正则化理论和凸集投影理论的基础上，利用比特流中的量化信息，提出了一种正则化投影超分辨率重建算法；通过正则化代价函数引入图像序列的时间域和空间域的先验信息，使用迭代梯度下降算法对正则化代价函数求解得到重建图像，最后利用凸集投影算法对求得的估计图像进行DCT域投影重建。仿真实验结果表明，该自适应算法较传统算法，其重建图像的主、客观质量有一定的提高，适合压缩图像的应用。