Abstract: This paper presents an improvement on the conventional active shape model(ASM) for face alignment. In this method, DCT coefficients of rectangular domain centering the lineament points are used in lineament matching, and a 2-D texture pattern search-and-fitting scheme instead of an 1-D analysis is used to improve the precision of the ASM method. DCT has good capability in abandoning correlation of data and concentrating energy of images. It withholds 95% energy of all by extracting only 3.52% DCT coefficients and also enhances the orientation speed. Experiment results demonstrate the efficacy of the method in locating facial features.

Key words: active shape model(ASM), active appearance model(AAM), principle component analysis(PCA), discrete Cosine transformation (DCT), face positioning

摘要： 在传统ASM模型的基础上，采用以轮廓点为中心的矩形区域的DCT系数进行轮廓匹配，提出了一种改进算法。该方法充分利用了轮廓点附近的二维纹理信息，改善了ASM方法采用轮廓点附近一维灰度信息导致模型匹配精度不高的问题。DCT 具有良好的去相关性能和能量集中性能，仅采取3.52%的DCT系数却平均保留了图像95%的能量，并采用八方向搜索最佳轮廓点，提高了定位速度。实验证明该方法非常有效。

关键词: 主动形状模型, 主动外观模型, 主成分分析, 离散余弦变换, 人脸定位

CLC Number: 

• TP391.41