Fusion of visual and infrared face biometric scores by projection

This paper presents a projection model to fuse the scores of a visual face verification system and an infrared face verification system. Essentially, the model consists of an arbitrarily number of linear projection vectors with randomly permuted elements. An equal error rate formulation is next adopted to learn the linear coefficients for projection. The learned model is consequently used for prediction of verification states of unseen data. Our empirical evaluation using a moderate size data set shows potential of the proposed method.