Hyperspectral Face Dataset Augmentation for Enhancing Face Recognition System Performance

Abstract

RGB images represent information using only three primary color channels: red, green, and blue. In contrast, hyperspectral imaging utilizes a broader spectrum to convey information, allowing hyperspectral face images to capture a wider range of details, thereby enabling a more accurate assessment of facial features and skin conditions. However, acquiring hyperspectral face images and building a substantial dataset is challenging due to the need for expensive cameras, sophisticated setups, and controlled environments. Currently available hyperspectral face datasets are limited in the number of subjects and samples, making it difficult to generalize face recognition systems that utilize hyperspectral imaging. To address these limitations, this paper proposes enhancing the diversity of hyperspectral face datasets through various augmentation techniques. Specifically, the image generation method StyleGAN2-ADA was employed to create new subjects and Face Pose Augmentation was used to generate various poses for the same subject, thereby increasing dataset diversity. Experiments were conducted based on the augmented hyperspectral face dataset to verify the impact of the proposed method on improving the performance of face recognition systems. Experimental results demonstrate that models applying the proposed augmentation method exhibit enhanced performance compared to previous models, making a significant contribution to the advancement of face recognition systems.