基于统计形状模型的面部特征提取随机方法

针对在复杂背景和部分遮挡情况下提取面部特征轮廓的困难，提出了一种基于统计模型的随机方法．该方法将面部特征轮廓作为动态随机过程的状态序列，并应用统计方法建立人脸整体形状模型和特征形状模型，分别构造面部特征间和面部特征内控制点样本的预测方程，最后利用序列蒙特卡洛方法估计随机状态．该方法给出了面部特征提取的随机描述，打破了确定性方法对单高斯分布和轮廓形状线性变化的依赖性，实现了轮廓的准确可靠提取．对100幅正面人脸图像的实验结果表明，轮廓定位相对误差仅为2．7％．对标准人脸检测数据库中传统算法很难处理的复杂背景和部分遮挡情况，该方法能够正确定位面部特征轮廓．

Probabilistic Approach to Facial Feature Extraction Based on Statistical Shape Models

A probabilistic approach to facial feature extraction based on statistical shape models is presented. Facial feature contours are regarded as configurations of a stochastic process driven by both a dynamics and a statistical data model. The dynamics that represent the state transitions within and between facial components is constructed based on 4 facial component models and a full face shape model, which are learnt from annotated face images. Finally, a sequential Monte-Carlo method is applied to estimate the contours. The proposed approach provides a probabilistic formulation for facial feature extraction and relaxes the linear and Gaussian assumptions in the traditional approaches, so that facial features can be obtained accurately and robustly. Leave-one-out validation test on 100 frontal face images shows that the average error of the proposed method is as low as 2.7%. Extensive experiments on Carnegie Mellon University face database show that desired facial feature contours can be obtained by the proposed method in the cases of cluttered background and partial occlusion.