Hyperspectral Data Processing: Algorithm Design and Analysis

V

Hyperspectral Information Compression

Due to enormous data volumes provided by hyperspectral imaging sensors via hundreds of contiguous spectral channels with high spectral interband correlation, it becomes increasingly evident that high compression ratios can be achieved for hyperspectral data without loss of significant information. By realizing tremendous benefits that can be gained by data compression, many efforts have been devoted to hyperspectral data compression. Since a hyperspectral image can be considered as an image cube, many hyperspectral data compression algorithms have been developed by taking advantage of existing two-dimensional (2D) image compression algorithms and extending these 2D techniques to their three-dimensional (3D) counterparts that are generally applied to video images. Most notable are JPEG2000 multicomponent and 3D SPHIT (Set Partition in Hierarchical Tree) that are extended from their counterparts, JPEG2000 and 2D SPHIT. However, there is a danger in taking such an approach without extra caution due to two unique issues encountered in hyperspectral images that never occur in pure pixel-based 3D images. One is the issue of subpixel targets whose size is smaller than pixel size/resolution. Such a target is generally embedded in a single pixel vector and their presence cannot be visualized by its spatial extent. In this case, pure pixel-based 3D image compression techniques may fail to capture their existence. The other is the issue of mixed pixel ...