Rate-distortion based adaptive distributed video coding.

摘要

Compression of digital videos has been a research topic for many years. A number of video compression techniques have been developed. Among all, distributed video coding systems are gaining attention, since it is desirable to design low-complexity video encoding systems that can result in reduced power consumption and cost for many recent applications, such as portable multimedia devices and wireless sensor networks.;This dissertation focuses on some advances of distributed video coding (DVC), namely bitplane selective decoding and adaptive quantization. The advances proposed in this dissertation target improving the rate-distortion and visual performance of the distributed video coding without jeopardizing the simplicity of the video encoder. Both of the advances proposed in this dissertation are based on an adaptive distributed video coding architecture. When dividing a video frame into several partitions, the system can allocate different bitrates to different partitions in order to better adapt to the local characteristics of the frame.;First, a bitplane selective decoding algorithm is proposed for pixel-domain DVC. The system is named BitpLAne SelecTive DVC (BLAST-DVC). In BLAST-DVC, the significance of each bitplane is measured at the decoder based on estimated distortion-rate ratios that make use of a correlation model for the original source information and the side information. Only the syndrome bits of the bitplanes that have estimated distortion-rate ratios higher than a target distortion-rate ratio are transmitted and are used to decode the associated bitplanes. The remaining bitplanes are estimated using a minimum-distance symbol reconstruction scheme.;Second, a transform-domain DVC system with rate-distortion based Adaptive QuanTization (AQT-DVC) is proposed. The transform-domain Wyner-Ziv frame is divided into partitions and is adaptively quantized based on estimated local rate-distortion characteristics for each partition. The rate-distortion estimation is performed based on a correlation model between the source information and the side information and can be applied at the decoder without adding complexity to the encoder.;In addition, an enhanced AQT-DVC is proposed in this dissertation to provide a more accurate estimation of the source correlation model. For this purpose, coarsely quantized DC coefficients are utilized to compensate for the inaccurate estimation caused by fast or non-linear motion.