The effect of expressions in a database of face images

This paper present the results of some face recognition experiments in which 3 different databases are used, each database giving a different representation of the same 100 people. The first database consists of 4 images per person, where each image shows a blank facial expression. The second database also contains 4 images per person, but each image shows a different facial expression: blank, smile, angry, and surprised. The third database is a combination of the other two databases and consists of 8 images per person. It is found that the database consisting of 4 different facial expression gives better classification results than the database with the 4 blank prototypes, and performs only slightly worse than the much larger database of 8 prototypes per person