Improvement of the reliability of human identification using mosaic facial image

This paper proposes a human identification system using mosaic facial images without being influenced by the lighting conditions or the reflection from eye glasses. The influence of the ambient fluorescent lighting is removed by taking the facial image in the near infrared band. The reflection from eye glasses is removed by utilizing polarized light. The size, the position and the inclination of the face are normalized by capturing the pupils and clipping the face on the basis of both pupil positions. In the identification, the illuminance of the face is normalized by using the correlation coefficient of mosaic data. As the result of the identification experiment using the facial images of 100 people, an identification rate of 100% was obtained in the case of 20*20 mosaic section number. When the pupil pair is extracted with a default threshold level of binarization, the identification time is within 1 second. It is made clear that the proposed system is effective for conceivable changes of position, direction and illuminance of a face in practical scenes.