Learning representative pose using eigenfaces

Face recognition from real world images is one of the most difficult tasks. Faces in real-world images have usually different poses from the known ones. This paper presents an empirical experiment for learning the most representative pose of human faces. Eigenfaces for each pose are determined and used to define the face space. Test images are projected into the face space and the prediction error rate is calculated. The experiment is implemented over seven different poses and different number of images. The results show that images with different sharp rotation produce different eigenfaces, and therefore, increases the error rate.