Hyper-Spectral Satellite Image Compression

by
Alex Cho

Hyper-spectral images consist of huge amounts of data divided into 224 separate bands. As in any situation that requires storage and transport of such data, efficient compression techniques are needed. We have explored various techniques that will save bandwidth with acceptable fidelity trade-offs. I will present one of the techniques that we have researched so far. The major component of this technique involves predicting one band of the image from another which increases the efficiency of current compression algorithms by a noticable factor. Unlike most situations, these images are compressed in one location, namely the satellite, and decompressed in another place, the Earth. The prediction ordering must be synchronized between both places while preserving the efficiency of compression. We used the minimum spanning tree algorithm for directed graphs to find the optimal ordering in which to compress the bands. I will present the algorithm in this talk.

Advised by Richard Ladner

MGH 228
Monday, February 3, 2003
3:30 - 4:20 pm