Cubic-panorama image dataset compression for image-based virtual environment navigation

This thesis addresses the key issues of image dataset compression, especially cubic-panorama image dataset compression, used in image-based virtual environment navigation to develop effective and efficient compression techniques and schemes. To my knowledge, this is the first work to be published to investigate and design compression schemes for cubic-panorama image datasets.;For cubic-panorama image dataset compression, a scalable lifted wavelet-based coding scheme with displacement compensation is developed. This scheme is based on the framework of lifted wavelet transforms with cross-image displacement-compensated enhancement combined with embedded entropy coding. The wavelet transforms efficiently generate hierarchically structured decomposition coefficients and provide the potential of spatially scalable coding required by image dataset compression. The lifting operations put wavelet analysis and synthesis into fast memory-saving in-place computations. The displacement compensation significantly improves the wavelet transform coding efficiency across basis images.;A new embedded entropy coding approach named independent block with layered data partition (IBLDP) is proposed. It combines embedded independent coefficient block coding with layered bitplane data partitioning. To provide the required spatial resolution scalability and spatial random access, no inter-subband and inter-block dependencies are exploited. Each subband is encoded independent of other subbands. With the compact layered bitplane data representation and the simplified coding structure, IBLDP features reduced computational and implemental complexity and hence is more suitable for the efficient and interactive image rendering application.;For cubic-panorama image dataset compression, a specific random access mechanism is designed. A new hierarchical data structure is proposed to accommodate the random access mechanism. The corresponding bit-stream syntax is formed to support this new data structure. Multi-level index tables are embedded in the bit-stream to easily facilitate the spatial image random access. A proper compromise between the coding efficiency and the random access flexibility is reached.;A spatially consistent representation of cubic-panorama image datasets is proposed. With this spatially consistent representation, unrestricted search for displacement vectors as well as matching reference blocks is extended beyond side image boundaries in all directions. A block padding algorithm for constructing reference blocks is presented for displacement estimation and compensation. Optimized matching reference blocks are obtained to reduce prediction errors and improve compression efficiency. Superior coding performance is achieved with the spatially consistent representation compared with the generic planar representation of cubic-panorama image datasets.;A novel global displacement estimation and compensation approach with scaled block-depth estimation is developed. Based on a six-parameter perspective projection model, a displacement-compensated image prediction algorithm is presented, taking into consideration all kinds of camera motions. A unique block-based scaled depth estimation technique is proposed for image prediction. The displacement-compensated predicting images are generated by using global model parameters combined with the estimated scaled block-depth map. More accurate predicting images with less-data presentations are obtained compared with the traditional BMA. The proposed approach is more efficient when applied for cubic-panorama image dataset compression.;Experimental results of the proposed techniques and schemes applied to encoding the testing cubic-panorama image datasets are demonstrated. Superior coding performances are achieved over those of the corresponding comparable techniques and schemes applied for cubic-panorama image dataset compression.