Combination of Face Direction Estimation and Face Recognition Using Four-Directional Features

To identify people’s faces from various directions, we propose a novel method that distinguishes people using a combination of face direction estimation and face recognition. Both the face direction estimation method and the face recognition method are appearance-based methods that use a linear discriminant analysis on the same features, the Four-Directional Features. Since these methods are constructed on the same framework, it is easy to combine them and extend the variable range of face directions. Our method consists of two hierarchical processes: the first is face direction estimation, and the second is face recognition. Estimating the face direction, we can use the face recognition discriminant space by limiting the range of face directions. Limiting the face direction can strengthen a face recognition discriminant space. Using the face recognition discriminant spaces extended the range of the face directions estimated, a total accuracy rate of face recognition was improved. Experiments showed that our method performed at an accuracy rate of 97.6 % for 105,000 images using 150 subjects and 35 different face directions in the range of ±45 degrees horizontally and ±30 degrees vertically.