Efficient processing of compressed images and video
Boon-Lock Yeo · Medical Entomology and Zoology · 1996
Image compression techniques have been developed for efficient storage and for efficient use of communication bandwidth. They are developed with little consideration to the image processing to be performed on the data. However, there is an increasing need for efficient processing and management of image and video information that is often in compressed formats. The processing of compressed data calls for the development of efficient techniques that operate on these data, as well as the development of compression methods that permit fast processing. In this thesis, a new framework is proposed to look at the processing of compressed images and video as an integral process for applications in video analysis and visualization of multi-dimensional data. Algorithms are developed to reconstruct reduced image sequences from compressed video, and to analyze the contents of the video based on the reduced images. Both exact and approximate reconstructions of reduced images from MPEG compressed video, with minimal decoding, are studied. Algorithms are developed to automatically segment the video and detect highlights on the reduced images, thus offering a significant reduction in computation while maintaining the effectiveness. Such analysis of video is an important step towards the automatic extraction of story units and highlights, and allows the nonlinear access of digital video for efficient browsing and navigation. Significant gains in terms of computation and resource utilization can be achieved by designing compression algorithms jointly with processing methods. A compression scheme is proposed which uses the knowledge of volume rendering to achieve visualization of compressed 3D scalar data without having to first decompress the entire dataset. The method allows volume rendering of localized data blocks, reduces complexity through homogeneity detection in transform domains, and permits more efficient and economical utilization of memory and storage. The various algorithms proposed in the thesis have demonstrated that data compression and image processing, when considered in combination, can offer significant gains in a variety of data-intensive applications.