Face Feature Localization And Face Modeling Based On Stereo Vision

Automatic face recognition is one of the most important subject in the field of pattern recognition. It can be widely used in the departments in charge of security, trading and transportation. There have been some deep researches in face recognition and showed quite reasonable results. But restricted by the resources in the face libraries, traditional 2D face recognition algorithms could hardly get rid of the impacts caused by light intensity, light direction and face pose. Because of the above problems, the recognition rate of the algorithms dropped dramatically and greatly restricted the use of the algorithms.In this thesis, we proposed an algorithm with the help of Digiclops stereo vision system. By means of 3D information of the face, we could get higher recognition rate. We use Digiclops Stereo Vision System to get color images and depth images of faces. Then we make use of skin luma and depth information to get the face area. Lastly we extract face contour in the face region using Snake algorithm; 3D data is necessary to benefit from independence of volume information relative to rotation and scale. A 3D description brings much more information, possibly rotation and scale independent. Nowadays with advance of 3D capturing tools, laser scanners and high-speed stereo machines, interests in 3D data processing have been increased. Because of the different advantages of 2D and 3D data, we tried to combine them together in order to extract the feature points of the face. Lastly we reconstructed 3D face models with 3 depth maps in different poses. We applied rotation and translation transformation to the 3D face models and projected it to the 2D image plane to get the 2D images of different poses. With the 2D images of different poses, we weakened impacts caused by different poses.