Score Fusion of SVD and DCT-RLDA for Face Recognition

• Main Authors: Bengherabi, M., Mezai, L., Harizi, F., Guessoum, A., Cheriet, M.
• Format: Conference Proceeding
• Language: English
• Subjects: Biometrics; Computational efficiency; Discrete Cosine Transform (DCT); Discrete cosine transforms; Face recognition; Image databases; Linear discriminant analysis; Matrix decomposition; Normalization; Regularized Linear Discriminant Analysis in DCT domain (DCT-RLDA); Score fusion; Singular value decomposition; Singular Value Decomposition (SVD); Spatial databases; Vectors
• ISSN: 2154-5111; 2154-512X
• DOI: 10.1109/IPTA.2008.4743776

Although information fusion in unimodal or multimodal biometric systems can be performed at various levels, integration of the matching score level is the most common approach. Starting from the fact; that the fusion will be efficient if and only if the fused approaches are complementary not fully competitive. We propose in this paper the fusion of two projection based face recognition algorithms: singular value decomposition (SVD) using the left and right singular vectors of the face image as a face feature stored in a matrix and regularized Linear Discriminant Analysis in DCT domain (DCT-RLDA) which is known by its computational efficiency in addition to discrimination power. Experiments conducted on the ORL database indicate that the application of the Min-Max, Z-score score normalization schemes followed by a simple fusion strategies (simple sum, weighted sum, append) confirm the benefits of the proposed approach in terms of identification rate and processing time.