A method of facial feature alignment based on multi-initialization ASM

Active Shape Model (ASM) is one of the most popular methods for facial feature alignment. However, because of the diversities of poses and expressions, the available methods of facial feature alignment based on ASM and their improvements are prone to get into the problem of local minima. In this paper, the facial feature location based on projection function is used to distinguish distinct poses, and a multi- initialization ASM method based on different poses is proposed. This method is able to avoid the ASM search process ending at local minima, especially efficient in dealing with face images which has diverse poses. The experimental results show that the method can enhance the accuracy of ASM with little increase of time requirement since ASM search time is reduced by further initialization.