Dual Variational Generation Framework for Enhanced Cross-Modal Face Recognition
M. Aimola Davies Anne, M. Brindha, N. Sivakumaran · 2024
One of the major challenges in face recognition is accurately identifying faces under varying lighting conditions, leading to a problem known as Heterogeneous Face Recognition (HFR). The objective is to integrate the capabilities of the modified Dual Variational Generator (DVG-Face) framework with traditional recognition methods, such as Local Binary Pattern Histogram (LBPH) and the HaarCascade framework, to provide a robust solution to the challenges of heterogeneous face recognition. This involves developing and implementing a model that successfully handles the complexity of HFR tasks utilizing synthetic images generated from Near-Infrared (NIR) and Visible (VIS) modalities. To guarantee identity consistency in the produced paired heterogeneous images, pairwise identity preserving loss is also used. The integration of Haar Cascades, the LBPH algorithm, and the DVG approach, along with person reidentification using the SSRL model, establishes a reliable framework for facial recognition in real-world security applications.