Abstract: This paper presents a new method to align the facial feature points, which is developed from the generalized procrustes analysis, and improved by adding nonlinear information of facial points and using the active shape model and the kernel trick.Experiments show that the improved method has better performances in obtaining more detailed modes of face shape and better accuracy in searching unknown target image points.Meanwhile less time is spent for the search.

Key words: active shape model, face recognition, face alignment, generalized procrustes analysis