针对超高清网络多媒体系统应用，遇到定码率模式下的视频压缩失真复原问题，以及视频压缩失真复原与超分辨重建割裂问题，拟通过压缩失真量化模型，DCT域学习模型，及多任务算法框架三个方面展开研究解决。首先，针对定码率模式下压缩质量的非均匀时空分布特性，基于编码信息对图像块进行重构，重点研究基于DCT域量化损失及预测编码准确性建模的压缩失真量化算法，及基于CNN的压缩失真评估算法。其次，基于DCT变换及反变换的线性可微特性，重点研究基于深度卷积神经网络与无监督学习的DCT域残差自适应估计方法。最后，基于视频图像压缩失真复原与超分辨重建的共性特点，研究灵活可变、低延迟、可流水的多任务视频图像修复算法框架，分别从效果最优与效率最优两个方面研究最优解决方案。最终目标是构建集视频压缩失真复原与超分辨率重建为一体的多任务视频修复算法框架，为超高清网络多媒体系统的发展提供核心技术支撑。

There are several problems for the application of UHD network-based multimedia system, including the video compression artifact reduction brought by fixed bitrate encoding, and the separation of video compression artifact reduction and super resolution. This project will be carried out on three aspects: compression artifact quantization model, DCT domain learning model, and multi-task algorithm architecture, to resolve these problems. Firstly, based on the nonhomogeneous distribution of the compressing quality brought by the fixed bitrate encoding, reconstruct encoded image with encoding information, then the research is focusing on DCT-domain quantization loss modeling and predictive encoding accuracy modeling based compression artifact quantization algorithm, and CNN-based compression artifact estimation algorithm. Secondly, based on the linearity and differentiability of DCT and inverse DCT, the research is focusing on deep convolutional neural network based and unsupervised learning based adaptive DCT-domain residual estimation algorithm. Finally, based on the common characteristics of video compression artifact reduction and super resolution, we focus on constructing a flexible, low-delay, and easy to be pipelined multi-task oriented video restoration algorithm architecture, and studying the optimal solutions respectively for effect oriented and efficiency oriented applications. The ultimate goal is to develop a multi-task oriented video restoration algorithm architecture for both compression artifact reduction and super resolution, providing the core technology support for the development of UHD network-based multimedia system.