Precise localization of facial keyponts, such as eye's centers, lips corners, nose tip, etc., is a necessary step for most face image analysis algorithms e.g. face recognition or classification. Also precise keypoint localization is necessary for image editing or manipulation. For example, augmented reality applications require keypoint localization to integrate virtual objects (hat or spectacles) in video stream.
In many cases for face normalization it is enough to supply only a few selected keypoints, e.g. corners or centers of eyes. However, to improve robustness and precision a more extended set of keypoints is required, which permits to take into account their reciprocal position.
Our face features detection algorithm permits localization of main anthropometric points in on-line video stream.