Abstract
When realizing a practical face identification system, higher robustness against fluctuation of the lighting condition is required. In this paper, a new face template-matching method dedicating to facial features (eye, nose and mouth) is proposed. It uses a special template for each facial feature, which consists of sub-divided region with compensated intensity. First, the size and incline of the input face image is normalized. Secondly, the facial features are extracted and separated into some small regions. Next, the intensity values in each region are compensated according to their average and variance. Finally, the correlation coefficients between the compensated features and templates are evaluated as an identification score. Experiments with face images taken in various lighting conditions prove that the proposed method is more robust than conventional methods based on the normalized correlation coefficient.
