This paper studies likelihood-ratio-based veri cation with an estimated, thus random, template. In particular, we have studied the case that N ≥ 1 feature vectors are available for template construction. The resulting, optimal, veri er is compared with two versions of the standard likelihood-ratio veri er: one with the expected feature vector μ as parameter, and one with μ replaced by an estimate μ_N.
展开▼