Image Compression for Precision Feature Extraction, Classification and Tracking

Practical methods for compressing image data will be developed by the EOS/Columbia University team for two applications that require highly reliable, object-based pattern recognition: (1) missile identification and tracking; and (2) multispectral image data archival and transmission for medical telediagnosis of wounds, burns and skin lesions. The compression technique must permit diagnosis by both human experts and machine-based ATR. Irreversible coding will be developed for real-time processing at rates higher than video. The reversible coding to be applied to lower-rate data will employ either block DCT or multiresolution DWT. Off-the-shelf digital hardware such as Datacube image processors will be used to implement the algorithms. In Phase 1, the compression methods will be tested both for the tracking of missile hard-body feature and for the automatic classification of skin lesions based on R, G, B image sequences representing clinical data. In Phase 2, the methods will be developed further and will be tested for real-time handover and aimpoint algorithms, and for early management of wounds and burns.