CASIA OpenIR  > 毕业生  > 硕士学位论文
基于形态学商图像和多向梯度二值特征的人脸识别方法研究
Alternative TitleMophological Quotient Image and Gradient Binary Pattern Based Face Recognition Methods
张瑶瑶
Subtype工学硕士
Thesis Advisor杨鑫
2008-06-04
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword人脸识别 光照变化 表情变化 商图像 局部特征 全局特征 Face Recognition Illumination Variation Expression Variation Quotient Image Local Feature Global Feather
Abstract随着社会科技的进步,人脸识别作为一种简便、安全、有效的身份认证方式受到了人们广泛的关注。但是由于实际应用中出现的光照、表情等的变化,极大的影响了人脸识别系统的性能。本文以人脸识别中的光照和表情问题为研究对象,分别从商图像和局部特征提取的角度对这两个问题进行的深入的研究,并且取得了一定的研究成果。本文的主要工作及贡献概述如下: 1. 提出了一种基于形态学商图像MQI的人脸识别方法,它是将形态学闭运算的方法引入到商图像中来。由于闭运算具有很好的边缘保持和尺度选择特性,因此它能够很好的进行光照估计。将图像转换到对数域上进行计算,可以加重图像中有用的特征信息所起的作用;为了减小结果图像中的噪声,采用了局部平滑的预处理手段;针对闭运算的尺度选择特性,我们根据图像的局部特征来自适应的选取结构元素的大小,这样可以更好的对光照进行估计。在Yale B和PIE人脸数据库上的实验表明,MQI方法能够很好的解决人脸识别中的光照问题,同时具有很低的时间复杂度。 2. 提出了一种基于多向梯度二值特征GBP的人脸识别方法,同时我们将其扩展到MGBP。该方法是通过对象素周围的一个很小的区域提取特征来计算属于该象素的标签,继而通过对人脸图像进行分割,统计每个分割出来的子区域的各个标签值的直方图,得到属于该子区域的局部特征表示。随后,将各个子区域的直方图按照一定的顺序串连起来,得到整幅人脸图像的全局特征表示,这种人脸的特征表示比当前流行的针对表情的特征表示具有更低的维数。在FERET大规模人脸数据库上的实验表面,该方法不仅具有较高的人脸识别准确率,而且在处理被噪声污染的图像时,仍然能够保持较好的识别稳定性。 总之,我们的方法分别在处理人脸识别中的光照和表情问题中取得了较好的效果,在保证较高的人脸识别准确率的同时,保持了较低的时间和空间复杂度。
Other AbstractAs the improvement of science and technology nowadays, face recognition, a kind of convenient, safe, and effective way of identity recognition, achieves more and more attention. Due to the unstable factors of illumination and expression in real application, there is a noticeable decline in the performance of face recognition system. In this article, we focus on the problems of the influence of illumination and expression in face recognition, and do some further research on base of Quotient Image and Local Feature Extraction. The main achievements of this article are as follows: 1. Mophological Quotient Image (MQI) based face recognition method is proposed to solve the problem of uneven illumination well. Closing operation is used to estimate the light condition. The face image is transformed to the logarithm domain, so that the key information can be promoted. In order to reduce the noise in the result, local-smoothing scheme is adopted. Furthermore, we choose the size of structure element dynamically to let it be adaptive to the local feature. Experiments on Yale B and PIE database show that such method can get a high recognition rate, and keep low computational cost as well. 2. Gradient Binary Pattern (GBP) is brought out to deal with the problem of expression variation, and we extend it to Middle & Gradient Binary Pattern (MGBP). The label of each pixel in the image is calculated. The face image is divided into several regions, and GBP histogram is extracted from each region. All the histograms are concatenated to form global feature to represent the face image, which has a lower dimension than most of current methods. Experiments on FERET database show that our method can be robust to the expression variation in face recognition, and keeps a quite stable performance when processing images with kinds of noise. In a word, our methods make a good performance to solve the problems of illumination and expression variation. Besides high face recognition rate, such methods have a low computational cost.
shelfnumXWLW1248
Other Identifier200528014628041
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7456
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
张瑶瑶. 基于形态学商图像和多向梯度二值特征的人脸识别方法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2008.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20052801462804(1919KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[张瑶瑶]'s Articles
Baidu academic
Similar articles in Baidu academic
[张瑶瑶]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[张瑶瑶]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.