基于ARG模型匹配算法的人脸特征定位

ARG Model Based Human Facial Feature Location

本文研究了基于 ＡＲＧ（ Ａｔｔｒｉｂｕｔｅｄ Ｒｅｌａｔｉｏｎ Ｇｒａｐｈ）模型匹配算法的人脸图象的特征定位方法。该方法可有效地完成眼、口、眉的自动定位。基于人脸生理特征我们建立人脸图象的ＡＲＧ模型对人脸图象进行了精确的描述。在此基础上本义采用了连续型Ｈｏｐｆｉｅｌｄ神经网络模型并行处理进行ＡＲＧ模型匹配，通过能量函数极小化得到全局最优的特征定位结果。实验结果证实了该方法有效、稳定、且具有较快的收敛速度。

A facial features location method based on ARG (Attributed Relation Graph) and a graph-matching method based on HF neural network are proposed. Here we can capture the contours of eye, mouth and eyebrow. First, based on physiological knowledge of human face we build thee ARG model. In the graph-matching step we proposed a HT algorithm and design a cost function. Experimental results with some real images show the method has good performance and high rate.