| Title |
Improving Face Recognition Performance through Hyperspectral Face Dataset Augmentation |
| Authors |
최영인(Youngin Choi) ; 강주성(Jusung Kang) ; 이지오(Jioh Lee) ; 신영학(Younghak Shin) ; 이흥노(Heung-No Lee) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.12.112 |
| Keywords |
Hyperspectral face recognition; StyleGAN2-ADA; Face pose augmentation; Data augmentation |
| Abstract |
RGB images use only three primary color channels to represent information, whereas hyperspectral images capture data using a broader spectrum. This allows hyperspectral face images to provide more detailed information, enabling a more accurate analysis of facial features and skin conditions. However, obtaining hyperspectral face images requires expensive cameras, precise equipment setup, and restricted environments, making it difficult to collect a sufficient amount of data. As a result, publicly available hyperspectral face datasets have limited objects and sample sizes, hindering the generalization of face recognition systems based on hyperspectral images. To address this issue, this paper proposes a data augmentation method to increase the diversity of hyperspectral face datasets. First, StyleGAN2-ADA, an image generation technique, was used to augment new facial objects. Next, Face Pose Augmentation was applied to generate diverse poses for the same object, further enhancing data variety. Through face recognition experiments based on hyperspectral images, the effectiveness of the proposed data augmentation method was validated across four models. The results showed that the models trained with the augmented hyperspectral face dataset exhibited improved accuracy, with performance increasing up to 0.9907. This demonstrates that the proposed approach can significantly enhance the performance of face recognition systems. |