Optimal Video Compression Parameter Tuning for Digital Video Broadcasting (DVB) using Deep Reinforcement Learning

Abstract

DVB (digital video broadcasting) has undergone an  enormous paradigm shift, especially through internet  streaming that utilizes multiple channels (i.e., secured  hypertext transfer protocols). However, due to the  limitations of the current communication network  infrastructure, video signals need to be compressed  before transmission. Whereas most recent research has  concentrated and focused on assessing video quality,  little to no study has worked on improving the  compression processes of digital video signals in  lightweight DVB setups. This study provides a video  compression strategy (DRL-VC) that employs deep  reinforcement learning for learning the suitable  parameters used in digital video signal compression. The problem is formulated as a multi-objective one,  considering the structural similarity index metric  (SSIM), the delay time, and the peak signal-to-noise  ratio (PSNR). Based on the findings of the experiments,  our proposed scheme increases bitrate savings while at  a constant PSNR. Results also show that our scheme  performs better than the benchmarked compression  schemes. Finally, the root means square error values  show a consistent rate across different video streams,  indicating the validity of our proposed compression  scheme. 

https://doi.org/10.59200/ICONIC.2022.030