一种改进的ASM人脸特征点定位方法

传统主动形状模型(active shape model, ASM)利用主成分分析(principal component analysis, PCA)获得形状，根据主成分并结合形状参数建立起统计模型，该方法不能有效更新图像特征空间，会导致模型匹配失败。为了克服上述缺点，提出一种改进的特征提取方法。首先采用增量主成分分析(incremental principal component analysis，IPCA)更新图像集合的特征空间，并采用一种新的模型匹配算法定位目标特征点。改进的新方法不断更新图像特征空间且有效地描述训练样本间的相似性或差异性，新的匹配算法能避免计算协方差矩阵的逆矩阵，能有效降低计算误差。实验结果表明，改进的方法有效提高了定位精度，并降低了匹配的像素偏移值。

Improved ASM localization method for human facial features

In original ASM model, the principal component analysis (PCA) approach is used to extract shape eigenvectors of the training data to build statistical model, which will lead to a failure matching of ASM because it can''t update feature space. In order to overcome the flaw, an improved ASM method is proposed. Firstly, the feature space is constantly updated by using the incremental principal component analysis (IPCA), and a new matching algorithm is proposed to locate facial feature. The improved ASM can describe the similarity or difference among training image sets, and avoid the calculation of the covariance matrix of the inverse matrix to reduce the calculation error. Experimental results show that the improved method can improve the matching accuracy and reduce the pixel displacement of ASM matching.