Content-based and temporal-scalable video coding

Abstract

In the design of video coding systems, compression efficiency and scalability are two of the most important topics under investigation. While channel bandwidth is still the most precious resource for many data networks, we require video coding algorithms that allow very high compression rate, yet retain reasonably good visual quality. To achieve this, the concept of content-based video coding is proposed recently to address the need of considering the features of video content to improve compression efficiency. On the other hand, scalability is another important aspect for video coding as it allows adaptation of video quality to different constraints incurred in the video transmission or retrieval processes. In this thesis, we firstly investigate the content-based video coding algorithms for low bit rate networks. The process of content-based video coding is divided into three parts: (i) object segmentation; (ii) feature extraction; (iii) adaptation to video coding systems. To demonstrate the approach, we apply the idea to the design of a content-based H.263 video codec for real-time coding of road traffic videos. Based on the proposed approach, moving objects (i.e. cars) are first segmented and classified based on their velocity. They are then coded in different frame rates accordingly. As compared with the conventional H.263 encoder using for the same application, the proposed system has a 20% increase in compression rate with negligible visual distortion. The proposed system fully complies with the ITU H.263 standard hence the encoded bit stream is completely comprehensible to the conventional H.263 decoder. As for the scalable video coding, a new temporal-scalable video codec is proposed in this thesis to provide compressed videos at different frame rates. The proposed codec is developed based on the interpolating wavelet transform. It shares the same advantage of the traditional temporal subband (TSB) approach in that its structure is very simple since it does not require the complicated motion compensation process. It outperforms substantially the TSB approach in generating lower frame rate videos.