請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/47034
標題: | 指向性局部直方圖均衡特徵及其在人臉辨識上的應用 Oriented Local Histogram Equalization Features and Its Application to Face Recognition |
作者: | Szu-Wei Wu 吳思蔚 |
指導教授: | 洪一平(Yi-Ping Hung) |
關鍵字: | 臉部特徵碼,稀疏重現分類,人臉識別, FTC,SRC,LBP,AR,FERET,Face Recognition, |
出版年 : | 2010 |
學位: | 碩士 |
摘要: | 在這篇論文中,我們提出了一個新的影像前處理方法藉由指向性局部直方圖均衡強化人臉圖片中局部具方向性的對比,並應用在人臉識別上。此方法主要透過在人臉圖像上使用不對稱濾窗作局部直方圖均衡,保留具方向性的局部資訊。為了擷取出人臉影像的特徵,我們串接八個不同方向指向性局部直方圖均衡的結果。因為特徵具有局部方向性資訊,固將其運用在人臉識別上,我們預期會有正面的效果,實驗上也證實了我們的推論。此方法主要的貢獻包含計算複雜度低、在不同照明下的不變性以及能簡單的和其他人臉識別演算法整合。我們示範了如何將其透過降維應用在全臉識別上,如稀疏重現分類(SRC)及局部人臉識別,如臉部特徵碼(FTC)。另外,我們提出了一個結合臉部特徵碼(FTC)及稀疏重現分類(SRC)的方法,名為稀疏重現臉部特徵碼(SRFTC)。此新方法的結合了兩個優異人臉演算法的優點並有效降低其缺點所造成的影響。由AR database上實驗的結果得知,全臉識別達到99.3%的辨識率,局部人臉識別上達到99.8%的辨識率。 In this paper, we propose a novel image preprocessing method which enhances local oriented contrast of facial images by using oriented local histogram equalization (OLHE), and apply it on face recognition. This method preserves local oriented information by performing local histogram equalization (LHE) with asymmetric kernels. In order to extract the feature on a facial image, we concatenate results which are processed by using eight different orientations of OLHE, called 8-oriented OLHE feature. We expect that the result of face recognitions will be better, because the feature contains local information and orientations, and our inference is proved by the experimental results. The key advantages of the method are its less computational complexity, invariance on illumination changes and can be integrated easily with other face recognition algorithms. We demonstrate the integrations of OLHE with Sparse Representation-based Classification (SRC) which is a holistic face recognition algorithm, and Facial Trait Code (FTC) which is part-based face recognition algorithm. Furthermore, we propose Sparse Representation Facial Trait Code (SRFTC) which is an integration of the FTC and SRC. This novel method combines advantages of these two algorithms and decreases the influence of shorts effectively. Based on the experiments on AR database, we obtain 99.3% recognition rate on holistic face recognition algorithm and 99.8% recognition rate on part-based face recognition algorithm. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/47034 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-99-1.pdf 目前未授權公開取用 | 1.96 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。