Full scalability supported video coding for ubiquitous video consumption

Abstract

In this dissertation, a full scalability supported video coding for ubiquitous video consumption is proposed. SVC targets at giving temporal, spatial, and SNR scalabilities to the encoded bitstream. In addition to these basic scalabilities, video has other scalabilities in the point of semantics. Thus, by enabling semantic scalability could help the consumption of video in ubiquitous environment. But supporting scalability decreases the coding efficiency, and the main reason in the decrease of coding efficiency is layered coding structure to support spatial scalability in SVC. Therefore, it needs to enlarge the dimension of scalability with minimal decrease of coding efficiency. In this thesis, a way how to use semantic scalability with basic scalabilities to extend the granularity and dynamic range of bitrate variation and enhance subjective quality is exploited, thereby full scalability is enabled in SVC. Supports of three semantic scalabilities such as Region of Interest (ROI), video summery, subjective quality in SVC are developed. Ways to describe multiple ROIs by FMO (Flexible Macroblock Ordering) are proposed. And constrained motion estimation, handling of fractional sample interpolation and upsampling for interlayer intra prediction in the ROI boundary are also proposed to support ROI scalability for independent ROI decoding. And to extract ROI bitstream from the whole video bitstream, an extraction procedure and related metadata for the procedure are also proposed. And to reduce the decrease of the coding efficiency by layered coding structure, improved interlayer motion prediction is also proposed. For inter-layer motion prediction, the use of refined motion data in the Fine Granular Scalability (FGS) layer is proposed instead of the conventional use of motion data in the base quality layer to reduce the inter-layer redundancy efficiently. The proposed methods are implemented in JSVM. The support of semantic scalabilities in SVC is verified b...