Improved Fingerphoto Verification System Using Multi-scale Second Order Local Structures

Today's high-end smartphones are embedded with advanced fingerprint biometric recognition systems that require dedicated sensors to capture the fingerprint data. The inclusion of such sensors helps in achieving better biometric performance and hence can enable various applications that demand reliable identity verification. However, fingerphoto recognition systems have some inherent advantages over fingerprint recognition such as no latent fingerprints, and it enables the possibility to capture multiple samples at once from a biometric instance with minimal user interaction. Thus, user authentication based on fingerphotos could be a useful alternative as we can re-use the smartphone camera to capture the fingerphotos. On the other hand, such an approach introduces different challenges; for example illumination, orientation, background variation, and focus resulting in lower biometric performance. In this research, we propose a novel verification framework based on the feature extracted from the eigenvalues of convolved images using multi-scale second order Gaussian derivatives. The proposed framework is used to authenticate individuals based on images/videos of their fingers captured using the built-in smartphone cameras. When combining with the commercial off the shelf (COTS) system, the proposed feature extraction technique has achieved the improved verification performance with an equal error rate of 2.76%.