Facial feature point extraction using a new improved Active Shape Model

In this paper, we present a new improved Active Shape Model (ASM) for facial feature points extraction. ASM performance is often influenced by some factors such as initial position, illumination, pose, etc, which frequently lead to the local minima in optimization. The original ASM has two sub-models: global shape model and local texture model, we proposed two improved methods for the shortcomings. First, we use a new method to obtain the pupil center position accurately, and these centers can provide more accurate initial position for the point distribution model of ASM. Second, we establish two texture submodels, and they constitute the new texture model together with the original: one based on YCrCb space and color similarity, and the other based on the non-skin region texture feature in the normal direction of facial feature point. The new method of pupil center location can make the average shape transform into initial shape and most similar to the actual feature points model; the new texture model can locate feature points precisely, it can make points located in the contour well. Our experiments of the proposed method have shown effectiveness comparing with the conventional.