Creating realistic images of virtual environments has long been a goal in computer graphics. The complexity in geometry-based modeling and rendering of real-world environments has led to the development of image-based representations such as light fields. Using a collection of multi-view images instead of geometric primitives, the light field representation bypasses the difficult and sometimes impossible problem of acquiring accurate geometry and illumination models. Rendering from a light field provides a simple and efficient way to generate arbitrary new views of the scene with photo-realistic quality, which requires only modest computational resources with rendering time independent of scene complexity.; A major challenge in rendering light fields is the large amount of data which must be compressed for efficient manipulation. In addition, the rendering algorithm requires quick access to small data segments at arbitrary locations within the compressed data. Tree-structured vector quantization (TSVQ) has made it possible to render directly from compressed data with its low decoding complexity and support for fast random access. However, TSVQ little exploits the coherence between images and thus provides only moderate compression, so that large, high-quality light fields can be viewed only on higher-end computers with large amounts of memory.; An analysis of the light field data access pattern during rendering reveals that, despite their high compression efficiency, video compression schemes such as MPEG or H.263 are not suitable for this application. A new compression algorithm, hierarchical disparity-compensated vector quantization (HDVQ), has been designed to address the problem of retrieving and decoding small data segments on the fly from arbitrary locations in a compressed data stream. This algorithm is shown to provide a compression ratio of as much as 10 times that of TSVQ, and comparable compression to H.263 when followed by a variable-length coder. With a small software cache, a HDVQ-based light field renderer achieves interactive frame rates, more than half of the frame rates obtained by a TSVQ-based renderer. By offering a much higher compression ratio with a small sacrifice in speed, HDVQ allows average users to enjoy interactive viewing of high-quality light fields on their home computers.
展开▼