An Encoding Scheme for Robust Face Recognition

Noticing that many available view-based methods are confined within rather limited operating conditions, in this work we Investigate Into whether external factors such as illumination, pose, expression and occlusion can be explicitly estimated and be used to boost the verification performance. Our approach consists of three steps: encoding the external factors, code estimation, and identity verification using the encoded data. Empirical experiments were performed on this method using the BERC database which contains both visual and infrared facial images. The experimental results show significant performance elevation from the proposed method, thus demonstrating the effectiveness of isolating the external factors.