Video Compression by Chroma Prediction Using Semantic Communications
Prabhath Samarathunga, Yasith Ganearachchi, Anil Fernando · 2024
Conventional video coding is evolving to meet unprecedented consumer device requirements, but the statistical signal processing based approach may find limitations in handling new media contents. Deep neural network and semantic communication based video compression systems show potential to be used as video encoders and decoders, but reaching the rate distortion performance of state-of-the-art conventional video coding systems remains to be achieved. A novel video compression system by predicting the chroma components of video using the semantically encoded luma component and reference intra-coded frames is proposed and tested against high-efficiency video coding for bit rate comparison and rate-distortion performance evaluation. The proposed system demonstrated 18% to 30% saving of bit rate for high and medium motion videos without significant reductions of rate distortion, the saving increasing with a larger group of picture sizes, but low motion videos only demonstrated negligible savings.