Estimation of Face Orientation from Shading Images
Kaori Suzuki, Hideo Saitô, Shinji Ozawa · IEEJ Transactions on Electronics Information and Systems · 1997
Sensing of human face orientation is vitally important tasks for applications to face to face teleconferences and so on. Most of methods for estimating face orientation from image processing demand extracting face features. However, feature extraction is much affected by environment inclusive of lighting and background, and some of them need manual operations. This paper proposes a robust method for estimating face orientation from shading images without feature extraction. First we obtain the 3-D shape and a shading image of object face by the equipment incorporated with a FG range sensor. Then head images with various postures are synthesized in advance. Next we compare an object image with synthesized images. The head-posture of the synthesized image having highest correlation with the object image is determined as the estimated posture. As a result of experiments using image sequences as the input, synthesized image sequences changed the face orientations following the change of face orientaions of input image sequences.