Scalable 3-D wavelet video coding

• Main Author: Zong, Wenbo
• Other Authors: Ma Kai Kuang
• Format: Theses and Dissertations
• Language: English
• Published: 2008
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing; DRNTU::Engineering::Computer science and engineering::Data::Coding and information theory
• Online Access: https://hdl.handle.net/10356/13120
• Institution: Nanyang Technological University

Concerning the various applications that involve heterogeneous networks, computing capabilities, and/or display terminals, such as Internet video streaming and live video surveillance, a scalable video coding system provides an effective solution to address this issue by offering some scalable functionalities including temporal scalability, spatial scalability and signal-to-noise ratio (SNR) scalability, while achieving high compression performance. Conventional hybrid coders like MPEG-2 and MPEG-4 are not able to offer scalability efficiently due to the recursive motion compensated (MC) prediction loop. In recent years, motion-compensated wavelet coding (MCWC) has emerged as a promising video coding approach to offering both high compression performance and versatile scalabilities. In this research, we first introduce the 3-D MCWC technology and its three possible structures: t+2D, 2D+t, and 2D+t+2D. We then investigate two issues related to MCWC. The first issue is the spatial scalability of the t+2D scheme, which suffers from degraded rate-distortion (R-D) performance for lower spatial resolutions. We investigate the problem from both the decoder side and the encoder side. At the decoder side, we present an in-depth analysis on the inverse motion-compensated temporal filtering (MCTF), or IMCTF, when reconstructing lower spatial resolution videos, and propose a method to improve the IMCTF process at the decoder. At the encoder side, we analyze the MCTF process in detail, and highlight the root source of aliasing that is not cancellable for lower spatial resolutions at the decoder. We propose a scheme named low-to-high lifting MCTF (LTH-MCTF) to eliminate this aliasing. Furthermore, we propose a practical solution, named pyramidal t+2D, that applies t+2D coding to each level of the frame pyramids to produce one separate bitstream with optimal R-D performance for each spatial resolution.