Face Recognition under Variable Illuminations

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

NSC96-2115-M019-001

Translated Name/計畫中文名

不同照度下的人臉辨識研究

Project Coordinator/計畫主持人

Mong-Shu Lee

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=1418928

Year

2007

Start date/計畫起

01-08-2007

Expected Completion/計畫迄

31-07-2008

Bugetid/研究經費

487千元

ResearchField/研究領域

數學

在此計畫中，我們提出一個在不同照度下的人臉辨識新方法，雖然主要成分分析(PCA) 與線性區別分析(LDA)在人臉鑑識工作上是廣為所用的工具，但每當臉部外觀受到一些因素如光線、表情及姿態變化時，其辨識率就下降很多，在我們的方法中，針對主要成分分析法中，處理固有空間投影步驟前，我們先利用發展出來的對偶樹小波轉換為基礎的結構相似度量測，來選取人臉影像的代表性固有向量，此構想的重點是想找出對照度變化較不敏感的人臉影像固有向量。先前的實作可看出此法在人臉辨識上提供較好的代表性且有較低的錯誤率。 In this project we proposed a new approach to face recognition under various illuminations. Although the principal component analysis (PCA) and linear discriminant analysis (LDA) are widely used tools for face discrimination tasks, their performance drop whenever face appearance are subjects to variation by factors such as illumination, expression and pose. In our proposed work, prior to the eigenspace projection like PCA, we employ the dual-tree wavelet transform based structure similarity (DTWT-SSIM) metric to select the representative eigenvectors of the face image. The thrust of the idea is to search for the image eigenvectors that are insensitive to the variations in illuminations. Preliminary experiments suggest that this new method provides better representations and achieves lower error rates for face recognition.

Keyword(s)

小波
人臉辨識
主要成分分析
Wavelets
Face recognition
Principal component analysis

DSpace CRIS

Face Recognition under Variable Illuminations

Details

Description