Facial Features Location And Tracking For Stereoscopic Display

With the development of face-detection technology, people put forward a new target of detection technology, which is facial features location and can track features in all angles. Facial features localization is an important research issue in the field of machine vision, involved in Machine Learning and Pattern Classification. This technology can also analyze the expression of viewer then give the viewer a better experience of viewing.This paper describes important machine-learning algorithms implementing an interactive three-dimensional display system and the main algorithm mainly includes following parts:The first part introduces the background detection and the way to choose candidate regions. The second part introduces the research on facial features localization, especially pupil positioning. The third part provides us tracking algorithms. We can solve the problem caused by rotation of the face.In order to predict the candidate regions, the paper proposes two methods: The aim of first method is to obtain the meaningful face region from the whole image by using skin color detection. The second is based on inter-frame difference and differential projection.Face detection algorithm is based on Adaboost and SVM[2]. Then we do some research on regress analysis of SVM and the algorithm does good performance in face detection.The key of facial features localization algorithm is ASM(active shape model). The facial features will transform into an ordered vector by establishment of contour model and texture model. Then we reduce the time of training by PCA. The last, templates are used to match and verify. The ASM algorithm is more accurate positioning comparing the haar.In order to make the system capable of real-time tracking and reduce the deformation caused by the rotation of the face, we provide three kinds of tracking algorithms named Kalman, Camshift. Finally, we solve the rotation problem by using tracking.