An Improved Neural Network Approach to End-to-end Video Compression
Liangwei Fu, Ping Wang, Xinhong Wang · 2024
The increase of network broadband and video users brings extremely severe challenges to the storage and transmission of massive video data. Therefore, enhancing the efficiency of video data compression has become a paramount concern. To solve these problems, an improved neural network approach to end-to-end video compression is introduced in accordance with existing algorithms. A lightweight preprocessing module and a more advanced optical flow generation network for motion estimation are introduced. Additionally, a simplified attention mechanism module is included in the encoding and decoding process of residual information. Furthermore, a reconstructive frame quality enhancement network module based on convolutional neural networks is integrated. Finally, all neural network modules jointly use an extended loss function. The empirical outcomes obtained from testing on benchmark datasets demonstrate that the proposed method surpasses the traditional video compression schemes widely used in academia and industry in terms of compression performance while also exhibiting improvements over mainstream end-to-end methods.