Eickeler, S.S.Eickeler2022-03-092022-03-092002https://publica.fraunhofer.de/handle/publica/34120110.1109/AFGR.2002.1004133This paper explores the face database retrieval capabilities of a face recognition system based on hidden Markov models (HMMs). A new HMM-based measure to rank images of the database is presented. The method is able to work on a large database. Previous systems for image retrieval based on HMMs were only capable of operating on small databases. The relation of the presented method to confidence measures is pointed out, and five different approximations of the confidence for the task of database retrieval are evaluated. The experiments are carried out on a database of 25,000 different face images, showing that the normalization and filler models are most suitable for retrieval on a large face database.enface database retrievalpseudo-2D hidden Markov modelface recognition systemdatabase image rankinglarge databaseconfidence measureconfidence approximationnormalization modelfiller model005006629conference paper