The naive representations of information, including an image represented as a set of pixels, or human thought written in languages based on alphabets, often are highly redundant; in addition, they contain information that the intended recipient cannot process, e.g., recording frequencies outside the human audible spectrum. Such representations can be compressed through information encodings that reduce redundancy or remove unneeded accuracy. In some instances, further reductions can be obtained by introducing small, inconsequential changes in the information to be encoded, i.e., distortion, or pixel values that differ from the original ones. This project is currently focusing on image compression and seeks to develop, study, and assess both lossless and lossy compression algorithms.
This year we developed tools for the selection of optimal orthogonal wavelets (biorthogonal wavelet generator), investigated new combinations of known compression techniques (development of code-books applicable to multiple image classes), and initiated joint (NIST-industrial partners) inquiries into the services needed to support effective video delivery (chiefly, adaptations of Forward Error Correction Codes that exploit the structure of the encoded video, and the fact that the constituent parts of the signal are not equally important and image estimation techniques to reduce the impact of signal loss).