Translator Disclaimer
21 February 2001 Multiscale representations for face recognition
Author Affiliations +
Proceedings Volume 4232, Enabling Technologies for Law Enforcement and Security; (2001)
Event: Enabling Technologies for Law Enforcement, 2000, Boston, MA, United States
Automatic face recognition algorithms have the potential to impact several applications including surveillance, human augmentation, multimedia indexing and retrieval, and authentication. In this paper, a technique to retrieve images by visual appearance similarity is applied to the problem of face recognition. The framework for representing and computing similarity is based on the design and use of multi-scale Gaussian differential features (MGDFs) as appearance features. In the first part of this paper, the relevance of MGDFs as appearance features and an algorithm to deduce global similarity is developed. In the second part of this paper, multi-scale representations are applied to face recognition. Results from experiments on standard test collections tested in this paper indicate that at least 96% recognition accuracy is obtained, and when compared with other techniques, the MGDF based representation yields comparable or better results. The MGDF based technique is very general; it was originally developed for global appearance similarity retrieval in heterogeneous images, and has been applied to retrieve similar textures, trademarks, binary shapes and heterogeneous gray-level collections.
© (2001) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
S. Chandu Ravela "Multiscale representations for face recognition", Proc. SPIE 4232, Enabling Technologies for Law Enforcement and Security, (21 February 2001);

Back to Top