In the case of using one-to-N authentication, the user identification means 105 compares the user to be authenticated and N registered users. The user identification means 105 calculates the distance between the feature value of the user to be authenticated and the feature value of each of the N registered users, and determines that a registered user with the shortest distance is the user to be authenticated. The user identification means 105 may use one-to-one authentication and one-to-N authentication in combination. In this case, the user identification means 105 may perform one-to-N authentication to extract a registered user with the shortest distance, and then perform one-to-one authentication using the extracted registered user for comparison. The calculated distance measure may be, but not limited to, Euclid distance, cosine distance, or the like.

Although the above describes an example where the feature value storage means 106 stores feature values obtained from a plurality of persons beforehand, the feature value storage means 106 may store a statistical model instead of feature values. For example, the statistical model may be a mean and a variance yielded from feature values acquired for each user a plurality of times, or a relational expression calculated using such a mean and variance. The statistical model may be a Gaussian mixture model (GMM) described in PTL 1, a support vector machine (SVM), a model using a neural network, or the like.