AI Based Biometrics Recognition with a Hyperspectral Image Sensor

In information security, facial recognition technology is one of the most familiar authentication technologies in our daily life. However, the recent development in the field of Artificial Intelligence has made it easy to generate a human face image from a small number of images, which poses a threat to individual information security. This is known as deep faking, and is a problem for future information security. As a solution for this issue, we develop a face recognition technique based on highdimensional images using a hyperspectral sensor (HSI). Unlike ordinary two-dimensional sensors, the high-dimensional data acquired here is a huge cube-shaped data with spectral wavelength information. In order to perform authentication using highdimensional images, we generated images for each wavelength band by dividing the wavelength information into pseudo-bands and created classification models for each band. Inference is performed on a single high-dimensional image using multiple trained classification models, and authentication is based on the majority vote of the models. We confirmed the effectiveness of these methods through validation experiments.