Timothy Sperr

Abstract

The H.264/AVC video coding standard leverages advanced compression methods to provide a significant increase in performance over previous CODECs in terms of picture quality, bitrate, and flexibility. The specification itself provides several profiles and levels that allow customization through the use of various advanced features. In addition to these features, several new video coding techniques have been developed since the standard's inception. One such technique known as Region of Interest (RoI) coding has been in existence since before H.264's formalization, and several means of implementing RoI coding in H.264 have been proposed. Region of Interest coding operates under the assumption that one or more regions of a sequence have higher priority than the rest of the video. One goal of RoI coding is to provide a decrease in bitrate without significant loss of perceptual quality, and this is particularly applicable to low complexity environments, if the proper implementation is used. Furthermore, RoI coding may allow for enhanced error resilience in the selected regions if desired, making RoI suitable for both low-bitrate and error-prone scenarios. The goal of this thesis project was to examine H.264 Region of Interest coding as it applies to such scenarios. A modified version of the H.264 JM Reference Software was created in which all non-Baseline profile features were removed. Six low-complexity RoI coding techniques, three targeting rate control and three targeting error resilience, were selected for implementation. Error and distortion modeling tools were created to enhance the quality of experimental data. Results were gathered by varying a range of coding parameters including frame size, target bitrate, and macroblock error rates. Methods were then examined based on their rate-distortion curves, ability to achieve target bitrates accurately, and per-region distortions where applicable.

Library of Congress Subject Headings

Digital video--Standards; Video compression--Standards; Optical pattern recognition

Publication Date

7-1-2011

Document Type

Thesis

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Kwasinski, Andres

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TK6680.5 .S74 2011

Campus

RIT – Main Campus

Download

Share

COinS