A Method to Enhance Face Biometric Security

In cancelable biometrics, a template undergoes a random-like transformation process such that it is disposable and a new template for the same identity can be re-generated for new usage. However, existing works do not consider control of verification performance and identity concealment at the same time. In this paper, we propose a method which can boost the verification performance of face biometric via two steps: using an efficient feature extraction transformation, and using an error minimizing template transformation. An extended Random Projection of face data improves the feature extraction efficiency in the first step. A near optimal transformation of the face template is next generated from iterative steps of gradient descent algorithm based on an error rate formulation. For experimentation, we adopted face images from AR and BERC databases where the face images were collected under various external conditions. From these face databases, we show an improved verification performance in terms of test equal error rate (EER), at the same time hiding the face identity based on template transformation. This work contributes to establishment of an algorithm for effective cancelable face template generation as well as verification performance enhancement.