Information measures for object recognition accommodating signature variability

This paper presents measures characterizing the information content of remo te observations of ground scenes imaged via optical and infrared sensors, O bject recognition is posed in the context of deformable templates; the spec ial Euclidean group is used to accommodate geometric variation of object po se. Principal component analysis of object signatures is used to represent and efficiently accommodate variation in object signature due to changes in the thermal state of the object surface. Mutual information measures, whic h are independent of the recognition system, are calculated quantifying bot h the information gain due to remote observations of the scene and the info rmation loss due to signature variability. Signature model mismatch is quan titatively examined using the Kullback-Leibler divergence. Expressions are derived quadratically approximating the posterior conditional entropy on th e orthogonal group for high signal-to-noise ratio. It is demonstrated that quadratic modules accurately characterize the posterior entropy for broad r anges of signal-to-noise ratio, Information gain in multiple-sensor scenari os is quantified, and it is demonstrated that the cost of signature uncerta inty for the Comanche series of FLIR images collected by the U.S. Army Nigh t Vision Electronic Sensors Directorate is approximately 0.8 bits with an a ssociated near doubling of mean-squared error uncertainty in pose.