We propose a low-complexity audio-visual person authentication framework based on multiple features and multiple nearest-neighbor classifiers, which instead of a single template uses a set of codebooks or collection of templates. Several novel highly-discriminatory speech and face image features are introduced along with a novel "text-conditioned" speaker recognition approach. Powered by discriminative scoring and a novel fusion method, the proposed MCCN method delivers not only excellent performance (0% EER) but also a significant separation between the scores of client and imposters as observed on trials run on a unique multilingual 120-user audio-visual biometric database created for this research. © Springer-Verlag Berlin Heidelberg 2007.
CITATION STYLE
Das, A. (2007). Audio visual person authentication by multiple nearest neighbor classifiers. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 4642 LNCS, pp. 1114–1123). Springer Verlag. https://doi.org/10.1007/978-3-540-74549-5_116
Mendeley helps you to discover research relevant for your work.