基于动态图像序列的表情识别  

Series of Dynamic Image Sequences Based on Expression Recognition

作　　者：贾熹滨[1] 闻春城 包锡元[1] 

机构地区：[1]北京工业大学计算机学院,北京100124

出　　处：《北京工业大学学报》2013年第9期1360-1365,共6页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(61070117);北京市自然科学基金资助项目(4122004)

摘　　要：为提高表情表述能力,提出建立组合单帧表情空域特征的表情序列联合特征.在分析Gabor小波的不同方向和尺度组合对表情图像表征能力基础上,确定采用3个方向和2个尺度的Gabor滤波器组提取单帧表情图像特征,描述表情动作的空域特征.在此基础上,组合连续表情图像序列的特征,建立包含表情动作变化过程的联合特征,解决了利用表情相关的局部空域和时序变化信息建立表情表述模型问题.利用支持向量机(SVM)作为分类器分别在JAFFE静态表情数据库和Binghamton动态表情数据库上进行测试,结果验证了静态图像采用Gabor+PCA特征比PCA特征更具有效性,表明利用动态表情序列建立表情特征比用静态表情图像具有更高的表情识别正确率.To improve the representability of emotion, a joint feature for expression sequences by combining spatial features of single expression image frames was proposed. On the basis of analyzing the representability for identifying expressions with the different combination of rotations and scales to Gabor wavelet, the Gabor filter with three rotations and two scales was adopted to obtain the static image feature. By connecting the feature of series expression images, the joint feature was established by containing the dynamic property of an emotion action. It solved the problem of expression description by using the expression relative local spatial information and the sequential change clues together. Support Vector Machine （SVM） was adopted as the classifier, and the test was done on the JAFFE static expression corpus and Binghamton dynamic expression corpus. Experiments prove the effectiveness of feature with Gabor and PCA comparing to PCA only. Results also show that the joint feature based on dynamic image sequences improve the expression recognition rate referring to static feature.

关 键 词：表情识别 表情特征 特征分析 表情变化序列 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]