Using infrared to improve face recognition of individuals with highly pigmented skin

• DOI: 10.1016/j.isci.2023.107039
• Publication Date: 2023-06-08T06:35:29Z
• Authors: Alex G. Muthua; Rensu P. Theart; Marthinus J. Booysen
• Keywords: Electronic Nose; Feature (linguistics)

Face recognition is widely used for security and access control. Its performance limited when working with highly pigmented skin tones due to training bias caused by the under-representation of darker-skinned individuals in existing datasets fact that darker absorbs more light therefore reflects less discernible detail visible spectrum. To improve performance, this work incorporated infrared (IR) spectrum, which perceived electronic sensors. We augmented images captured using visible, IR, full spectra fine-tuned face systems compare these three. found a marked improvement accuracy AUC values receiver operating characteristic (ROC) curves including IR increasing from 97.5% 99.0% faces. Different facial orientations narrow cropping also improved nose region was most important feature recognition.