基于相位一致性图像的模块化PCA人脸识别方法

Novel modular PCA method based on phase congruency images for face recognition

张秋余 靳艳峰 袁占亭

提出了一种新的基于相位一致性的模块化PCA的人脸识别方法。解决了人脸识别受光照影响的问题。首先得到人脸训练样本的相位一致性图像;然后将人脸相位一致性图像划分为更小的子模块,用PCA方法处理这些子模块图像。在姿势、光照以及表情变化的情况下同一个人的局部面部特征是不变的,因此用该方法来处理这些变化。给出了传统的模块化PCA方法与该方法在不同姿势、光照和表情变化条件下的对比实验结果。实验结果表明该方法的人脸识别率较传统模块化PCA方法有了较大提高。[著者文摘]

A novel modular PCA algorithm for face recognition based on phase congruency was presented. The accuracy of the modular PCA method and the proposed method were evaluated under the conditions of varying expression, illumination and pose using standard face databases. The results indicate high improvement in the classification performance compared to the conventional modular PCA method.[著者文摘]

较少标定点条件下的主动形状模型研究

Research on active shape model under fewer landmarks conditions

冀贞海[1,2] 邹采荣[2] 赵力[2]

主动形状模型（ASM）的核心部分包括两个子模型，全局形状模型和局部纹理模型，两者均由标定人脸的许多关键点经过统计方法构建而成。本文研究了在只有少量关键标定点情况下，如何有效的进行主动形状建模的问题。通过充分利用面部轮廓标定点的信息，提出了一种将面部轮廓质心动态变化和传统主动形状模型相结合的策略，解决了在少量标定点条件下主动形状建模所遇到的问题。实验表明，改进方法在定位人脸图像关键点时耗时更短，而且具有更广范围的搜索能力和性能良好的搜索精度。[著者文摘]

The Active Shape Model （ASM） method is composed of two parts, global shape model and local texture model, which are all derived from the sufficient landmarks around the shape. This paper presents an improved method on how to utilize the ASM efficiently under fewer landmarks conditions. Through changing the gravity center of face contour dynamically and combining with the traditional ASM, the improved method overcomes the problem of modeling ASM under fewer landmarks. The experiments show that the improved method consumes less time in searching target shape than traditional ASM, and has slight wider capture range and same good performances in point location accuracy.[著者文摘]

基于粗略到精细分类的面部表情识别方法

Method of Facial Expression Recognition Based on Coarse-to-fine Classification

崔洁 冯晓毅

为了更准确地识别人的表情，在识别人脸7种基本表情（愤怒、厌恶、恐惧、高兴、无表情、悲伤和惊讶）时，采用了局域二值模式技术提取面部特征，进行由粗略到精细的表情分类。在粗略分类阶段，7种基本表情中的2种表情被选为初步分类结果（候选表情）。在精细分类阶段，选用计算加权卡方值确定最终分类结果。采用日本的Jaffe情数据库来验证算法性能，对陌生人表情的识别率为77．9％，其结果优于采用同样数据库的其他方法，且易于实现。[著者文摘]

This paper presents an effective method for facial expression recognition, It analyzes seven basic expressions： angry, disgust, fear, happiness, neutral, sadness and surprise. The local binary pattern（LBP） operator is used to extract face appearance features. A two-stage classification method is proposed. At the first （coarse classification） stage, two expression candidates from initial seven are selected. At the second （fine classification） stage, one of the two candidate classes is verified as final expression class by the Chi square statistic. Our algorithm is tested on the Jaffe database and the recognition rate for a new person＇s expression is 77.9%, which is higher than all the other reported methods.[著者文摘]

基于GA的2DPCA在人脸识别中的应用

Application of 2DPCA in face recognition based on GA

孙瑞霞 汪亚明 黄文清

采用基于遗传算法（GA）的二维主成分分析法（2DPCA）进行人脸识别。2DPCA直接以二维图像矩阵为研究对象,以其协方差矩阵的特征向量为投影轴进行特征提取。为了达到识别时的信息最优,将遗传算法融入2DPCA,对协方差矩阵的特征向量进行优化选择得到最优投影轴,并在此基础上提取特征。最后在MIT人脸数据库上进行实验,表明识别率和速度均高于单纯使用2DPCA的方法。[著者文摘]

A method for automatic face recognition using genetic-algorithm based 2DPCA is proposed. Using a group of eigenvectors of the image matrix＇s covariance matrix as its projection axis, 2DPCA withdraws features directly regarding two-dimensional image matrix as its study object. In order to get an optimal solution, GA is applied to select projection axis from the eigenvectors of the covariance matrix. Finally, some face recognition experiments using MIT face database are given, which showed the performance of the new method is better than that of 2DPCA,[著者文摘]

交互式进化计算在虚拟角色表情建模中的应用

Application of interactive evolutionary computation in virtual actor＇s expression modeling

武春友 王士同

阐述了一种将交互式进化计算应用于三维计算机动画影片中训练虚拟角色面部表情的方法。将模糊神经网络应用于虚拟角色面部表情的建模，并用遗传算法结合交互式进化计算训练模糊神经网络。实验结果表明，当该算法应用于训练虚拟角色面部表情这一问题时，能在保证精度的前提下快速收敛，并能避免陷入局部最优，从而使得不同的动画导演利用这个系统，能够生成符合自己要求的虚拟面部表情。[著者文摘]

An Interactive Evolutionary Computation （IEC） method was illustrated for training the facial expressions of the virtual actors in three dimensional computer graphical movies according to the effect that the human directors expect. Fuzzy neural network was used to accommodate director＇s expert knowledge so that appropriate virtual actor＇s facial expression could be produced. In order to speed up the convergence of the learning algorithm, genetic algorithm was incorporated into the fuzzy neural network with IEC. Experimental results indicate the success of our approach here.[著者文摘]