Wavelet-based scalable coding of still and time-varying stereoscopic imagery

Abstract

This thesis addresses the issue of encoding and decoding still and time-varying stereoscopic imagery. A review of current encoding techniques is undertaken, with special emphasis on algorithms having SNR and spatial scalability. A stereo image pair consists of two views of the same scene. Due to the redundant nature of both views, prediction-based techniques produce superior results when compared with independent encoding of both images. Some of the most widely used embedded still-image coding techniques rely on discrete wavelet transform (DWT)-based analysis. However, these schemes cannot be adapted in a straightforward manner to encode stereoscopic still-image pairs. In this thesis, a novel DWT-based embedded stereoscopic still-image codec structure is proposed. This scheme preserves the progressive transmission capability of still-image coding algorithms, while suitably adapting to the nuances and special characteristics of stereoscopic imagery. A comparative study of variable-block and fixed-block disparity estimation is also undertaken. Partition artifacts result due to imperfect disparity compensation. Drawbacks in existing compensation techniques are discussed and a novel loop-filtering scheme is proposed. This is used to smooth disparity-compensated images before generating and subsequently encoding residual images. As seen from this thesis, this scheme improves on the performance of current techniques. In addition, the dyadic sampling structure of a 2-D DWT is exploited to obtain discrete levels of spatial-scalability and forms part of an embedded scheme for transmission of stereoscopic still-images at different spatial resolutions. The proposed algorithm is suitably modified to encode time-varying stereoscopic imagery. Drawbacks of current moving-picture hierarchies are analyzed and a novel hierarchy is proposed that insures that a user has the flexibility to view a sequence either in monoscopic (default) or stereoscopic modes. Independent objective results, explaining SNR and spatial scalability features, are presented when encoding a few pictures of a stereoscopic moving image sequence. In addition, informal subjective results are presented when viewing encoded versions a time-varying sequence.