Objective: Development of wavelet-based parallel data compression algorithms for data browsing with Earth and space science (ESS) imagery databases.
Approach: Our research involves developing parallel data compression algorithms using wavelet transforms. We have developed wavelet transforms that remove the "block" structures when reconstructed after data compression and have implemented the transforms on different parallel platforms. In doing so, we also evaluate which parallel hardware platform gives the best unit cost per operation for wavelet-based data compression. For image compression purposes, some of the desktop parallel hardware platforms with limited capabilities (e.g., CNAPS from Adaptive Solutions Inc.) perform equally well and yield a much lower unit cost per operation when compared with other high-end platforms (e.g., the CRAY Y-MP).
Significance: Our work addresses the problem of data compression for very large ESS imagery using wavelet transforms so that the compressed data can be used for initial query and browsing by the user.
Status/Plan: We have implemented the wavelet transform on several parallel platforms, including the CRAY Y-MP, MasPar MP-2, and CNAPS. We are now developing vector quantization (VQ) methods that will be used for compression of the wavelet transformed images. We have improved the wavelet transform algorithms on the MP-2 and obtained a factor of 10 improvement over the existing algorithms based on "cut and stack" virtualization. A paper based on this work is being presented at the Fifth IEEE International High Performance Computing Conference, 10/95. Plans include porting the VQ to the MP-2 and developing a highly parallel wavelet implementation for the "hierarchical" virtualization of the MP-2.
Points of Contact:
Srini Raghavan
Radha Poovendran
Image Understanding Group, LNK Corporation
raghavan@tove.cs.umd.edu or srini@zeus.lnk.com
radha@nibbles.gsfc.nasa.gov