摘要
提出了一种结合深度学习方法的GRU神经网络模型;通过采用CNN和GRU神经网络结构,将连续若干帧的CTU图像信息依次通过CNN和GRU结构中,训练学习视频空间和时间内容相关性,预测每一帧的编码单元分割结果;通过对基本视频测试序列压缩的实验验证和对比表明,提出的方法大多能有效降低编解码复杂率。
A GRU(Gated Recurrent Unit)neural network model combined with deep learning method was proposed.By adopting the CNN and GRU neural network structure,the CTU image information of several consecutive frames was sequentially passed through the CNN and GRU structures,and the learning video space and temporal content correlation were trained to predict the coding unit segmentation result of each frame.The experimental verification and comparison of the basic video test sequence compression show that the proposed method can effectively reduce the complexity rate in most cases.
作者
任学超
薛红平
武彦君
王学才
平建创
REN Xuechao;XUE Hongping;WU Yanjun;WANG Xuecai;PING Jianchuang(Northern Institute of Automatic Control Technology,Taiyuan 030006,China)
出处
《兵器装备工程学报》
CAS
北大核心
2019年第5期142-145,共4页
Journal of Ordnance Equipment Engineering
