基于统计形状分析的人脸基本表情分析

Facial Expression Recognition Using Statistical Shape Analysis

人脸的表情识别分析在虚拟交互和情感计算中具有重要意义,为得到一个精确的表情分析模型,采用统计形状分析方法对人脸表情进行建模分析,可将人脸表情的形状看作随机变量并在高维空间中服从一定的概率分布的.首先使用统计形状分析方法对表情形状进行归一化后,采用PCA方法对表情的变化建立点分布模型;在表情的分析阶段采用高斯分布模型估计表情空间的统计参数,并采用贝叶斯分类实验验证分类效果.实验结果表明:统计形状分析方法可以有效地对人脸的基本表情进行分析分类.

Facial expression interpretation and recognition is a key issue in visual communication and human-machine interaction. This paper proposes a face expression analysis method based on the statistical shape analysis in order to achieve an accurate model of facial expression. Facial feature contours are regarded as configurations of a stochastic process governed by a statistical shape model. After computing the mean shape and aligning all shapes resulting from the training set by means of a Procrustes analysis, shape variations are estimated using the Principal Component Analysis. Then, the set of facial expression shapes are modeled using a multivariate Gaussian distribution and maximum likelihood estimation is applied to the data from the models. The conditional probability and the prophetic probability of the shape data are estimated. Finally, an experiment to classify some basic facial expression is set up, demonstrating that the proposed method works with adequate efficiency and can recognize a given number of facial expressions.