Research Of Facial Landmark Localization Based On Statistical Learning

Face alignment is a key issue in the field of facial image analysis and recognition research. Many facial image related works, such as face recognition, facial expression recognition, pose estimation, human age estimation, human-computer interaction,3D facial animation modeling and etc., are depended on the accurate localization of facial landmarks. However, due to the imaging conditions in real world, most of the facial images suffer from many effects, such as light variations, pose variations, expression variations as well as some partial occlusions caused by hair, hands, glasses or other objects. All these uncontrolled effects make the face alignment become a challenging problem. The accuracy and robustness of existing algorithms still cannot meet the requirements of practical applications.In this thesis, some new methods are studied aiming at the landmark localization problem for faces with complex variability. The main contributions are summarized as follows:(1) Aiming at the face alignment in video frames, a facial landmark tracking method is proposed by adding Lucas-Kanade optical flow constraint on AAM fitting. The optical flow tracks some significant facial feature points, which are used for com-puting the inter-frame correspondence. The current initial shape for AAM fitting is then estimated by the similarity preservation between frames. Experiments show that the proposed method can successfully track the frames with which AAM tracks failed. The proposed method achieves not only the accuracy but also the computation time improvement.(2) A two-step face alignment approach using statistical models is proposed. AAM is known to be sensitive to initial values and not robust under inconstant circumstances. In order to strengthen the ability of AAM performance, a two-step approach for face alignment combining AAM and ASM is proposed. In the first step, AAM is used to locate the inner landmarks of the face. In the second step, the extended ASM is used to locate the outer landmarks of the face under the constraint of the estimated inner landmarks by AAM. The two kinds of landmarks are then combined together to form the whole facial landmarks. To improve the performance of ASM, a 2D local texture model is proposed to replace the 1D local intensity vector model. Experimental results show that the proposed approach gives a much more effective performance.(3) A robust face alignment method using classified random ferns and pose-based initialization is proposed. Compared with localization methods using deformable mod-els, the cascaded regression models obtain much better performance in terms of accura-cy and efficiency for those unseen images. A new random fern constructing method by introducing classification analysis as well as a robust pose-based initialization method is proposed by computing the pose similarity between the estimated image and initial shapes. The proposed method is evaluated on two challenging databases, LFPW and HELEN. The experimental results have demonstrated that the proposed method can significantly enhance the accuracy and robustness on facial landmark localization.