Abstract: Through three-dimension (3D) transformation shape search on two-dimension (2D) face image, the variation of the face gesture can be simulated correctly, and the nonlinear variations of face on Active Shape Model (ASM) are solved effectively. Based on standard face 3D model, the three-dimension coordinates were gotten by 2D initial shape; and 10 gesture-parameters of 3D transformation were calculated through the iterating process of 3D transformation, projecting to track the object shape. The test data show the 3D transformation shape search produces better results on searching multi-gesture face images that are not included in ASM training set.

Key words: 3D transform, Active Shape Model (ASM), feature-point, shape search, face gesture

摘要： 通过三维变换的特征点搜索方法对二维人脸图像进行形状搜索，形状搜索以人脸的标准三维模型为基础，将二维坐标三维化；通过三维变换、投影等过程迭代搜索逼近目标形状，最后得到三维变换的10个姿态参数。测试结果表明：在对非活动形状模型(ASM)训练集中的多姿态人脸进行形状搜索时，三维变换方法具有更好的逼近效果。较好地模拟了人脸姿态的实际变化，有效地解决了ASM人脸偏转的非线性问题。

关键词: 三维变换, 活动形状模型, 特征点, 形状搜索, 人脸姿态