Abstract: This paper proposes a novel face alignment algorithm: improved Bayesian Tangent Shape Model(BTSM). It isolates gesture parameters and shape parameters, deals with these two parameters respectively and uses radius of curvature as the number of dynamic search of this landmark. Experiments show that the algorithm can make up for the defects that BTSM can not well handle the alignment of multi-gesture face images and the approach can improve the effectiveness in both speed and accuracy.

Key words: face alignment, face recognition, Bayesian Tangent Shape Model(BTSM)

摘要： 论述一种新的人脸对齐的算法——改进的贝叶斯切线模型。该算法从形状参数中分离出姿态参数和身份属性参数，对这2个参数分别处理，通过引入曲率半径作为该特征点搜索的像素个数实现动态搜索。实验表明，该算法可以弥补BTSM不能很好处理多姿态的人脸图像对齐的缺陷，提高搜索精度，减少搜索时间。

关键词: 人脸对齐, 人脸识别, 贝叶斯切线模型

CLC Number: 

• TP311