Radar target identification using an eigen-image approach

In order to maintain a high probability of correct classification the classifier must provide good separation between target classes and must be robust with respect to target variability. The authors have implemented a new target classifier based upon the eigen-image concept developed by Turk and Pentland (see Journal of Cognitive Neuroscience, vol.3, no.1, 1991) for automatic recognition of human faces. This paper describes their new eigen-image classifier and presents preliminary performance results for a three-class (tank, APC, gun) classifier. Performance results are compared with those of a shift-invariant pattern matching classifier and a quadratic distance correlation classifier. The algorithms are compared by presenting classifier-performance confusion matrices, which indicate the probability of correct and incorrect target classification. The ability of each classifier to reject cultural false alarms (buildings, bridges, etc.) is also quantified