| Facial feature points location and points matching are hot fields in computer vision and pattern recognition. In the mean while, they are important perquisites for many applications such as facial recognition, facial animation, facial tracking, three-dimensional facial reconstruction, stereo matching and statistical model construction. Although the algorithms of facial feature points location are relatively mature on two-dimensional images, the research of the feature points matching on the three-dimensional model is still an open problem. In this paper, an algorithm for three-dimensional facial feature points matching is presented.On the three-dimensional point distribution model, the Relative Angle-Context Distribution (RAC) algorithm can locate the corresponding points between two models, but it can only track the corresponding points in a small area, therefore is not able to match the corresponding points of the facial model accurately. In order to improve the accuracy of feature points location, an algorithm called Facial Feature Points Location based on a Combined K-means clustering and Support Vector Machine was proposed which is based on the RAC clustering and Support Vector Machine. Firstly, RAC and K-means clustering method are combined in this algorithm by the way of searching pre-matches feature points, which are called the cluster point set, of the unknown model. Next, the geometric features of the model points such as the curvature, normal and the mean and variance of EE are extracted. Finally; cluster point sets are classified by SVM method to achieve the purpose of feature point separation and precise points location.Experimental results show that this algorithm can locate feature points more precisely on three-dimensional point distribution models than the RAC algorithm. When the distance threshold is 30, with the RAC algorithm,25%of the feature location accuracy was 100%and 50%of feature location accuracy was 100%with the proposed algorithm. |
PDF Full Text Request