The impact of lossy image compression on automatic target recognition performance

In distributed underwater signal processing for area surveillance and sanitization during regional conflicts, it is often necessary to transmit raw imagery data to a remote processing station for detection-report confirmation and more sophisticated automatic target recognition (ATR) processing. Because of the limited bandwidth available for wireless transmission, image compression is of paramount importance. Furthermore, it is equally crucial that image coding algorithms be evaluated according to some meaningful criteria. Instead of assessing the performance of image compression algorithms in terms of peak signal-to-noise ratio (PSNR) or normalized mean-squared error (NMSE), we resort to a more meaningful performance metric that reflects human and operational factors-ATR performance. We develop a novel image compression algorithm that achieves the minimal information state by a combination of subimage-specific transformation, principal component analysis, and vector quantization (VQ). We quantify the performance of image coding by extracting key parameters or features from the original and reconstructed images and by comparing the ATR performances using separate test sets that contain both mines and mine-like clutter. We achieve a compression ratio of up to 57:1 with minimal sacrifice in P _D and P_FA