一种姿态变化下的ASM人脸特征定位方法

ASM facial feature location under attitude changes

为了解决传统主动形状模型(ASM)在姿态变化情况下的人脸特征定位不准确的问题,提出一种姿态变化下的ASM人脸特征定位方法。首先在ASM初始化过程中添加一个旋转因子R来初始化人脸形状向量b,使得当人脸有姿态变化时,能自动调整人脸的偏转方向;然后融合局部二值模式算子建立局部纹理模型;最后在IMM人脸库上进行测试,平均定位误差为7.102 0个像素点,比传统的ASM提高35.85%。实验结果表明,与传统的ASM相比,该方法显著改善了姿态变化下的人脸特征定位不准确的问题,并提高了对各特征点的提取精度。

In order to solve the problem that ASM facial feature location under the attitude changes was not accurate,this paper proposed a new method of ASM facial feature location under the attitude changes.Firstly,it added a rotation factor R to initialize the shape vector b in the ASM initialization process,which automatically adjusted the face while the attitude changed.And then it integrated the local binary pattern operator to establish the local texture model.Last,it tested on the IMM database.The average positioning error was 7.102 0 pixels,increased by 35.85% than the traditional ASM.The experimental results show that the proposed method significantly improves the location accuracy of the traditional ASM under the attitude changes,and enhances the detection accuracy of all the feature points.