International Journal of Computer Science and Technology

International Journal of Computer Science and Technology | Vol. 8, No. 12, November 2006 | pp. 89–96

DOI: 10.46882/2006/IJCST/081227

Research Article

Title: Cross-View Face Recognition via Bilinear Subspace Projection Matrix Transforms

Names of Authors: Takahiro Tanaka¹, Jean-Pierre Dubois²

Authors’ Affiliations: ¹Department of Information Science, University of Tokyo, Tokyo, Japan; ²Laboratoire d'Informatique, Université de Paris-Sud, Orsay, France

Abstract:
Variations in facial pose present a significant challenge for automated biometric systems configured for unconstrained video surveillance environments. This paper develops a cross-view face recognition model based on a bilinear subspace projection matrix transform that maps arbitrary pose profiles into a unified canonical coordinate framework. The preprocessing system extracts regional shape and intensity descriptors before calculating view-invariant identity factors. The framework was evaluated using the multi-PIE database containing face vectors captured at up to 45-degree horizontal angles. Experimental outcomes achieved an overall cross-view facial verification accuracy rate of 94.25%, outperforming baseline principal component analysis models by 18.6%. The projection mapping computation completed execution within t = 16.2 ms per frame, confirming that this spatial transformation technique can support live multi-camera verification environments.

Keywords: Face Recognition, Pose Invariance, Subspace Projection, Bilinear Transform, Biometric Security

Manuscript Timeline: Received August 12, 2006; Revised September 24, 2006; Accepted October 15, 2006; Published November 22, 2006

Citation: Tanaka, T., & Dubois, J. P. (2006). Cross-View Face Recognition via Bilinear Subspace Projection Matrix Transforms. International Journal of Computer Science and Technology, 8(12), 89–96.

Subscribe to read the full article: https://www.internationalscholarsjournals.org/subscribe-to-read