微表情识别算法研究

发布时间：2018-04-13 20:14

本文选题：微表情识别 + 局部二值模式　；参考：《南京邮电大学》2017年硕士论文

【摘要】：表情识别作为人机交互的一个重要领域,已经得到了几十年的发展,在许多领域都有着广泛的应用。近年来,人们开始研究一种特殊的表情——微表情,微表情是一种持续时间短、强度弱、反应出一个人内心真实情感的特殊表情,其在测谎、临床诊断以及审讯等领域有着广泛的应用。本文针对微表情进行了相关的研究,使用了静态图像、动态序列以及深度学习的方法进行微表情识别,主要工作内容包括:(1)对微表情图像做预处理。本文所采用的数据库为CASME2和SMIC微表情数据,对数据库中的微表情图像做尺度归一化以及灰度归一化操作。(2)研究了基于静态图像的微表情识别,选出数据库中每一个样本的表情变化最大帧,当作该样本的静态微表情,提取局部二值模式(Local Binary Pattern,LBP)、局部相位量化(Local Phase Quantization,LPQ)特征并将其进行融合,实验结果显示,融合后的特征对微表情识别率有较大提高。(3)研究了基于动态序列的微表情识别,使用正交三维局部二值模式(Local Binary Pattern from Three Orthogonal Planes,LBP_TOP)算子提取动态序列的微表情特征,并使用局部线性嵌入算法(Locally Linear Embedding,LLE)算法对高维特征进行降维。LBP_TOP算子能够提取微表情在时间维上的信息,相较于静态图像的方法,其识别率更高。(4)研究了基于深度学习的微表情识别,将微表情序列输入3D-CNN网络中提取微表情特征,最后使用支持向量机(Support Vector Machine,SVM)进行分类,与其他深度学习方法相比,3D-CNN能够直接处理视频或者图像序列,计算简单效率高。
[Abstract]:As an important field of human-computer interaction, facial expression recognition has been developed for decades and has been widely used in many fields.In recent years, people have begun to study a special kind of emotion-microemoji, which is a kind of special expression which is short duration, weak intensity and reflects a person's inner true emotion.Clinical diagnosis and interrogation are widely used.In this paper, we do some research on microfacial expression. We use static image, dynamic sequence and depth learning method to recognize microfacial expression. The main work includes: 1) preprocessing microfacial expression image.The database used in this paper is CASME2 and SMIC microfacial expression data. The microfacial expression recognition based on static image is studied by scale normalization and grayscale normalization operation.The maximum frame of expression change of each sample in the database is selected, which is regarded as the static micro-expression of the sample. The local binary mode of local Binary pattern is extracted and the local phase quantization (LPQs) feature of Local Phase quantification is fused. The experimental results show that,The microfacial expression recognition based on dynamic sequence is studied by using the local Binary Pattern from Three Orthogonal operator of orthogonal three-dimensional local binary mode, and the microfacial feature of dynamic sequence is extracted by using the local Binary Pattern from Three Orthogonal operator.And the locally Linear embedding algorithm is used to reduce the dimension of high dimensional feature. LBP top operator can extract the information of micro expression in time dimension, compared with the method of still image.Its recognition rate is higher. (4) the micro-expression recognition based on deep learning is studied. The micro-expression sequence is input into 3D-CNN network to extract the micro-expression feature. Finally, support vector machine (SVM) is used to classify the microfacial expression.Compared with other depth learning methods, 3D-CNN can directly process video or image sequences, and the computation is simple and efficient.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18

【参考文献】