Efficient Real-Time Distributed Video Coding by Parallel Progressive Side Information Regeneration

Abstract

Distributed video coding (DVC) provides a different compression paradigm that consists of lightweight encoder and complex decoder. This compression paradigm shed a light for nowadays mobile devices which equipped with limited battery and computing resources while the corresponding multimedia service is backed up by cloud computing infrastructures. After years of developing, the coding performance and the decoding speed of DVC are still facing serious challenges, for example, DVC’s rate-distortion (RD) performance is far behind the traditional prediction based video codec (e.g., H.264) and the required decoding complexity handicaps its value in practical usage. In order to meet these challenges, we propose an efficient real-time DVC framework with progressive side information regeneration by taking GPGPU-based parallel computing into account. The proposed system can resolve one of the achilles’ heels, i.e., the large group of picture scenario, of DVC codec with good RD performance and much improved decoding speed. The Bjontegaard Delta peak signal-to-noise ratio improvement of our system is up to 2.87 dB as compared with DISCOVER codec, which is also comparable with that of the state-of-the-art optical flow-based DVC codecs. The average decoding speed goes up to 22.44 frames per second (FPS) and even up to 76.35 FPS for Hall sequence. To the best of our knowledge, this is the first DVC codec which tackles the RD performance and the decoding speed issues simultaneously.