抄録
In this paper, two fingerprint verification methods based on the discrete and continuous hidden markov model (HMM) are proposed, and those performances are evaluated. First, the optimal number of state for the continuous HMM is investigated. In case of x direction analysis, the number of state is less than the number of state which are necessary in case of y direction analysis, because there is a blank area in the upper part of fingerprint image. Next, the verification abilities of the discrete and continuous HMM are compared. When the HMM for one finger was built from 10 fingerprint images for a fingerprint registration, 100 [%] and 99.9 [%] verification rates were obtained respectively using the discrete and continuous HMM (Type II error 0 [%]). However the memory capacity which is needed for a fingerprint registration is fewer in the continuous HMM, and it is shown that about 6k bytes are necessary.
