摘要
针对静态表情特征缺乏时间信息,不能充分体现表情的细微变化,该文提出一种针对非特定人的动态表情识别方法:基于动态时间规整(Dynamic Time Warping,DTW)和主动外观模型(Active Appearance Model,AAM)的动态表情识别。首先采用基于局部梯度DT-CWT(Dual-Tree Complex Wavelet Transform)主方向模式(Dominant Direction Pattern,DDP)特征的DTW对表情序列进行规整。然后采用AAM定位出表情图像的66个特征点并进行跟踪,利用中性脸的特征点构建人脸几何模型,通过人脸几何模型的匹配克服不同人呈现表情的差异,并通过计算表情序列中相邻两帧图像对应特征点的位移获得表情的变化特征。最后采用最近邻分类器进行分类识别。在CK+库和实验室自建库HFUT-FE(He Fei University of Technology-Face Emotion)上的实验结果表明,所提算法具有较高的准确性。
To overcome the deficiency of static expression feature, which lacks time information and can not reflect the subtle changes of expression adequately, a dynamic expression recognition method is proposed for non-specific face： the dynamic expression recognition based on Dynamic Time Warping （DTW） and Active Appearance Model （AAM）. Firstly, the method of DTW based on local gradient Dual Tree-Complex Wavelet Transform （DT-CWT） dominant direction pattern is used to warp expression sequence. Secondly, using AAM to locate 66 feature points of face image and track them. The changing feature of expression can be obtained by calculating the displacement of corresponding feature points in two adjacent expression sequences image. And using the feature points of neutral face to build the facial geometry model. The matching of facial geometry model can overcome the expression differences between various people. Finally, the nearest neighbor classifier is used for classification and recognition. The experimental results on CK＋ database and HeFei University of Technology-Face Emotion （HFUT-FE） database show that the proposed algorithm has a high degree of accuracy.
出处
《电子与信息学报》
EI
CSCD
北大核心
2018年第2期338-345,共8页
Journal of Electronics & Information Technology
基金
国家自然科学基金(61300119
61432004)
安徽省自然科学基金(1408085MKL16)~~
关键词
动态表情识别
动态时间规整
主动外观模型
双树复小波变换
主方向模式
Dynamic expression recognition
Dynamic Time Warping （DTW）
Active Appearance Model （AAM）
Dual Tree-Complex Wavelet Transform （DT-CWT）
Dominant direction pattern
