文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > Facial expression recognition based on PCA reconstruction
侵权投诉

Facial expression recognition based on PCA reconstruction

基于PCA重建的人脸检测

The 5th International Conference onComputer Science & EducationHefei, China. August 24–27, 2010

WeP3.2

Facial Expression Recognition Based on

PCA Reconstruction

CHEN Juanjuan

Renai College of Tianjin University,

Tianjin, China, cjj8718@hotmail.com

Abstract—A facial expresion recognition algorithm named PCA (Principal Component Analysis) reconstruction to the still image containing facial expression information is proposed. The general idea of the algorithm is classifying the training sets of facial expression into different subsets by facial expresssssssssion. Then apply the PCA algorithm into the subsets to get the corresponding orthogonal basis. Test images project on the orthogonal basis of different expreion ubet, and recontruct baed on the projection coordinates. Then compare the tested image with the reconstructed images and choose the expression of the most similar one to be the expression of the tested image. The algorithm is tested on Japanese female facial expresion (JAFFE) database. The feasibility of the method has been verified by experiment.

ZHAO Zheng , SUN Han, ZHANG Gang

Tianjin University, Tianjin, China, zhengzh@tju.edu.cn

as much as possible, performing expression information as much as possible, we take measures to extract facial region which contains the most abundant facial expressions. These regions are eyes area and mouth area. By this way, we throw off face shape, the proportion of facial organs and much other personal information, and keep down the expression information at the same time. As the feature extracted by PCA may not be the best information for pattern recognition, the paper proposes an algorithm named PCA reconstruction.

II. PCA RECONSTRUCTION

The method is tested on JAFFE database. There are seven expressions of ten persons in the database including the expression happy, fear, sad, anger, disgust, surprise and Index Terms—PCA reconstruction, Affective computing, Facial

expression, Orthogonal basis. neutral. We divide the database into two parts. One is the

training set, and the other is the testing set. The general idea of

I. INTRODUCTION the algorithm is classifying the training sets into several

Affective computing [1] is a research field referring to different expression subsets by expression. In the JAFFE

database, that is, divide the training sets into seven subsets many subjects. It tries to let computers understand and express

their feelings like human, so it takes an important place in the corresponding to seven expressions and then apply the PCA intelligent human-computer interaction field [2]. Face method in the subsets. expression is an important way for people to express their Apply the PCA method in all the subsets and we obtain the feelings. Therefore, we must take facial expression recognition orthogonal basis. Because the PCA algorithm extracts the best research into the affective computing research and pay expression characteristics of the images, then the obtained attention to it. Facial expression recognition[3][4] is a rich orthogonal basis is the basis that expresses the expression research field in relation to artificial intelligence, computer space best. In other words, the expression images can be well vision[5], image processing, psychology, recognitions and so expressed by the linear combination of the basis. The test on. With the research of facial expression, we can effectively images project on the seven orthogonal basis which are got advance the progress of human machine interface system and from different subsets and then reconstruct the tested images. the automatic [6] understanding of computer images [7]. There are many differences between the original image and the One of the classical methods of facial expression recognition is PCA algorithm. Apply the PCA method to form the facial projection coordinates, but there is something not optimistic. The facial projection coordinates not only take the facial expression, but also take some information of people themselves, such as face shape, the proportion of facial organs and so on. These are why PCA applying in the research of face recognition, however, the personal information will impact the results of facial expression. Furthermore, by the PCA methods, we can find the best expression of training images sets but they may not be the best information for pattern recognition. The feature extracted by PCA is not the best expression category features. According to these affecting factors, with the aim of the extracted feature removing personal information

reconstructed image. The reason is the principle of PCA algorithm is reducing the dimensions of the image and expressing it as a linear combination of smaller form-based. These reductions are not lossless. The more we reduce the dimensions, the more differences there are. The PCA method is used on all the expression subsets, we get seven basis and the basis is corresponding to the subset. Images which are not belong to the subset project on the orthogonal basis, reconstruct the images and then get a reconstructed image. The reconstructed image is much more different from its former one. Images which are belong to the subsets project on the orthogonal basis, reconstruct the images and then get a reconstructed image. The reconstructed image is much more similar to its former one.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email