On estimating biometric capacity: An example based on LDV biometrics

It is known that when a pattern recognition system is not subject to memory constraints on pattern representations and sensory information is not compressed, the recognition rate is bounded by the mutual information between the memory and sensory representations. In this paper, we investigate the recognition rates of Laser Doppler Vibrometry (LDV) signals obtained from 285 individuals. We consider four cases corresponding to four different assumptions as to the structure of the data source and noisy measurements, and present the results of the mutual information bounds. In particular, we show the bounds of the recognition rates for each feature of the LDV signal.