Cross-Spectral Biometric Performance Analysis on High-Resolution Face Images
Pandian Shanmuganathan, Praveen Kumar
MetadataShow full item record
Biometrics is increasingly being used to authenticate the identity of individuals in critical use case devices like smart phones, laptops, and several other access-control systems in our day-to-day lives. Additionally, biometrics is also used in forensic and security sensitive areas to detect and identify suspects involved in bombings, robberies, and several police investigations. In each of these critical scenarios, the high-quality full-frontal face images of the subjects are not accurately captured as those subjects do not intend to register their identity to the Closed-Circuit Cameras. Hence in such cases, identification of the suspects becomes difficult even though we have the best recognition systems in place. The periocular region is the area around the eye including the eyebrows, eyelids, crow’s feet, iris, and the area under the eye. The main aim of this thesis is to analyze the performance of the periocular region when compared to the face and the iris modalities. The Face recognition is done using a commercial SDK. The iris recognition is done using VASIR, an open source Iris SDK. The periocular recognition is done with the help of support vector machines and histogram of gradients using a deep learning library, “D-L". The classifiers are trained separately with near-infrared, visible, and cross spectral images to analyze the performance on each spectrum. Though there are studies performed to analyze the cross spectral performance of the face, iris, and periocular regions separately, this will be the very first study to examine all the three modalities in different spectrums. This study would help in determining the significance of subject recognition in the NIR and Visible spectrum, where iris and face recognition systems typically fail. Upon completion of the experiments, it is evident that though iris and face are prominent biometric modalities, there are scenarios where the face and the iris recognition fails when the quality of the image is poor, as it does not meet the required threshold to qualify the recognition process. In such cases, the periocular region is still able to detect and identify the failed iris and face subject images. The periocular region is robust to some of the common variations like pose changes, occlusions, lighting, and illuminations when compared to the face or iris, therefore it can detect and recognize the subjects amidst of low quality unconstrained images.