High Compression of Infrared Data

ABSTRACT: Data compression schemes generally fall into two main categories: lossless and lossy. Lossless schemes promise to provide an uncompressed file which is identical before and after compression. As lossless schemes must be able to reconstruct every feature of the original file they often provide modest compression ratios. In fact, it is easy to show that no lossless compression scheme can compress all possible input files. On the other hand, lossy schemes have license to discard some measure of the original data and, as a result, they often result in much better compression ratios. However, due to the lossy nature of the compression algorithms, some information of the original data is omitted or distorted during the compression; therefore, the quality of the resulting file is often only controllable at the most rudimentary of levels. Is it possible to have good compression ratios typical of lossy schemes, while at the same time maintaining accuracy and rigorous control over data loss on a per pixel basis? As we will demonstrate in this proposal such algorithms are indeed possible and practical for real world problems. BENEFIT: Algorithms for performing advanced processing of EO/IR imagery have wide applicability in several domains including many viable transition paths within the Department of Defense ranging from missile defense to intelligence data gathering. Beyond military applications, there is a vast range of commercial applications such as spacecraft monitoring and video surveillance for security. The key idea of our work is the development of efficient compression schemes whose error is controlled on a per pixel basis. In standard lossy compression algorithms, some information of the original data is omitted or distorted during the compression; therefore, the quality of the resulting file is often only controllable at the most rudimentary of levels. Our methods, on the other hand, allow error control on a per pixel basis. Such methods allow the user of the data to have confidence that the compression scheme has not removed features of the data, nor introduced artifacts, that reduce the quality of imagery. Some of the ideas proposed here have already found footing in geospatial data compression, and these themes are positioned to have a similar impact in the IR intelligence domain at large.