FaceSensor-ID: A High Precision Approach on Multi-Spectral (Visible and NIR) Camera Sensor Origin Identification

The primary objective of this work is to enhance face recognition accuracy. While numerous approaches exist, this paper focuses on a specific area. We introduce FaceSensor-ID, a classification tool to identify the operational spectrum and select the most suitable face recognition matching model. This high-precision, deep learning-based method aims to determine the image sensor origin from a single visible or near-infrared (NIR) face image. FaceSensor- Id holds significant utility in forensic and biometric applications, enabling data tagging in multi-sensor face image datasets based on the originating imaging sensor. We test our method on five face image datasets, namely, CASIA-FaceV5, CASIA V2, Extended Yale B, TD IR E, and TD RGB E. These datasets include face images captured by six different imaging sensors, where 4 are visible and the other 2 are NIR sensors. To automatically identify the origin of image sensors, our proposed approach integrates image quality metrics with the utilization of a carefully chosen pre-trained model. Following hyperparameter tuning, we refine the model's performance. After cross-validation, the proposed model achieves nearly perfect accuracy across all aforementioned face image datasets.