Face Recognition Based on Fuzzy Connective Fusion of SVD and RWLDA Algorithms

Multimodal data fusion is a powerful methodology for improving the accuracy of biometric authentication systems that use modalities such as face, voice, and fingerprints. Essentially, there are two stages of data fusion in biometric systems: pre-classification fusion (prior to matching) and post-classification fusion (after matching). The primary objective of this research is to design and develop a post-classification methodology that combines scores from various dimensionality reduction algorithms for face recognition. To this end, we propose exploring a recent technique involving fuzzy fusion of normalized scores generated by Singular Value Decomposition (SVD), which is based on left and right singular vectors, and Relevance Weighted Linear Discriminant Analysis using QR decomposition (RWLDA/QR). This technique is inspired by fuzzy logic theory and employs an efficient linear combination of triangular norms (T-norms and T-conorms). Advanced simulations based on three complex databases—GT, AR, and LFW—were conducted to demonstrate the effectiveness of the proposed approach in terms of improving face recognition rates.