Surveillance Source Compression with Background Modeling for Video Big Data
Ling Tian, Hongyu Wang, Qinyu Tang, Yimin Zhou · 2016
Video source is a kind of critical component in big data. As a typical video source, the cost of storage and transmission is extremely high for surveillance source. Thus, it is the coding target to get a tradeoff between bit-rate and visual quality for surveillance video. Devoted to this subject, this work proposes a new background modeling scheme for surveillance source, which adopts the residual gradient and the block edge differences to construct background picture. The constructed background picture preserves the spatial characteristics of the source contents and then it is chosen as the long-term reference picture. This work proposes a novel background-based coding optimization algorithm (BCOA) for both picture level and the largest coding unit (LCU) level in video compression. According to the effective adjustment of quantization parameter (QP) and lagrange multiplier (λ), the proposed BCOA improves the visual quality. Compared with the background-modeling-based hierarchical prediction structure optimization, experimental results show that BCOA achieves better visual quality and BD-Rate gain up to 46.68%.