视频编码是解决视频数据存储和传输的有效手段，也是当前国内外研究的热点问题。然而，传统的视频编码技术大都采用全遍历暴力搜索最优编码参数和凭统计经验以手工设计预测函数的编码方案，这在很大程度上限制了视频编码性能的提升。本项目拟从计算机视觉和人工智能的角度出发，将信号处理领域问题模型化为人工智能领域问题，即将视频编码中的关键模块，预测编码、模式决策和运动补偿建模为图像修复、分类和超分辨率重建问题，利用神经网络为工具进行求解，同时探索编码失真环境下神经网络的攻击与防御，进而设计与训练对抗神经网络以应用于下一代视频编码标准，实现快速决策、自适应、智能化的高效编码，最终取得智能视频编码优化理论与方法的创新。

Video coding is an effective solution for the video storage and transmission, and it is also a hot research topic. However, brute force search for the optimal coding parameter and manually designed prediction functions are utilized in the traditional video coding, which greatly limits the improvement of video coding performance. From the perspective of computer vision and artificial intelligence, the key modules in video coding, including predictive coding, mode decision and motion compensation can be modeled as image inpainting, classification and super-resolution respectively, where the deep learning tool will be adopted. In addition, the attack and defense of neural network with coding distortion will be exploited to design and train the adversarial neural networks for versatile video coding, aims to achieve the theoretical innovation and technical breakthroughs of intelligent video coding.