Rapid data reduction and target detection in literal imagery

We present a method to perform target detection in EO imagery by rapidly reducing imagery data volume as algorithmic complexity increases. The technique is based on perceptual grouping of segmented regions through generic geometric, intensity and topology constraints that distinguish vehicles from the background. At each of the three processing stages, greater than 90% data reduction can be achieved, even at high false alarm rates. The typical compression ratio of the complete system can be very high while retaining virtually all significant information. The efficacy of the algorithm is demonstrated by experimental results on large reconnaissance images of wide-area battlefield scenes showing camouflaged vehicles in complex backgrounds.