Deep Spectral Biometrics: Overview and Open Issues

Spectral biometric systems record behavioral or physiological features, such as the face, iris or fingerprint, for identifying individuals. These systems function by acquiring images in various sub-bands of the electromagnetic spectrum. These systems recently gained traction for their use in applications such as defense, nighttime surveillance and airport security. Spectral biometric systems have shown promise since they are resistant to spoof attacks at sensor. Widely used biometric systems, work by acquiring images in the visible range of the electromagnetic spectrum. Such systems are examples of broadband imaging systems where information, in the captured images, is averaged over the visible spectrum. As a result, meaningful information is lost. On the other hand, spectral biometric systems acquire images in a different range such as Near-Infrared, Short-Wave Infrared or at a specific wavelength, retaining additional, useful information, beyond that which is recorded by human vision. This can then be exploited against vulnerabilities present in security systems. Further, this additional information can be used by machine learning/computer vision systems for robust personal identification. In this book chapter, we present recent advancements in spectral biometric systems, where deep learning based methods have been used to identify the biometric traits (face, iris and fingerprint) from spectral images. We also present an overview of spectral biometric systems and the limitations of deep learning based methods for such systems.