| Stereoscopic image display offers a simple and compact means of portraying on 2D screens the relative depth information in a real world scene. Perception of relative depth considerably enhances the viewing experience. Typically, more than two views would have to be transmitted either to provide the correct perspective to each viewer in a multi-viewer scenario or to provide a single viewer with the feel of "look-around". This results in a multi-fold increase in bandwidth over the existing monoscopic channel bandwidths. Achieving a significant reduction in the excess bandwidth needed for coding stereoscopic video, over the bandwidth required for independent coding of these multiple views, is the primary objective of this thesis. To this end, we present a framework for stereoscopic image sequence compression that brings together several computationally efficient algorithms in a unified fashion to address critical issues such as, (1) tailoring the excess bandwidth to be commensurate with the demand for stereoscopic video, (2) compatible coding (in terms of quality and technology), (3) scalability of coding efficiency and computational complexity with multiple views, and (4) synthesis of intermediate views to provide motion parallax perception to the viewer.; At the heart of the framework is a computationally efficient multiresolution and quadtree decomposition based segmentation scheme that provides an optimal representation for coding a stereoscopic image sequence by jointly minimizing the overhead for coding the segmentation structure and the segment disparities or displacements. Two configurations for stereoscopic sequence coding are investigated. In the first configuration, a complete disparity map is available at the decoder for each stereo-frame to facilitate synthesis of intermediate views. In the second configuration, the coding efficiency is improved by relaxing this restriction. We also present two segment-tracking-based joint coding approaches, whose coding efficiency and complexity scale well with multiple views. Methods for handling regions that are uncovered during segment tracking are outlined. Our framework also supports the psychophysically-motivated mixed-resolution-based coding which can lead to a significant reduction in the excess bandwidth without affecting the perceived stereoscopic quality. Over a test set of stereoscopic sequences the performances of our stereoscopic sequence coding extensions are presented and compared with the performance of two baseline schemes, one for each configuration, that are representative of the approach used by current international stereoscopic video coding standards. |
