An Improved Active Shape Model for Facial Feature Location

Active shape model (ASM) has been widely accepted as one of the best methods for image understanding. In this paper, we propose to improve ASM by introducing Procrustes analysis technique in the matching of feature landmark points of a set of training images and strengthening the edge in searching face profile. Firstly, each landmark point labeled manually is matched by its local profile in its current neighborhood. Then, by analyzing the variations of shape over the training set as in the ASM, we build a mean shape model (MSM) which can mimic this variation. The principle component analysis (PCA) is exploited in this part. Thirdly, we must adjust the parameters which can best fit a model instance to a new image and then the new image can be interpreted after much iteration at the end. Our experiments of the proposed method have shown some effectiveness comparing with the conventional ASM.