基於(2D)2-PCA+(2D)2-LDA之掌形辨識

Hand-Shape Recognition using (2D)2-PCA +(2D)2-LDA

戴君揚(Chung-Yang Tai)

指導教授 ： 陳文雄

國立暨南國際大學/科技學院/電機工程學系/碩士(2009年)

摘要

近年來社會上對安全問題日益重視，使得以生物辨識為基礎之身份辨識技術開始被大眾所重視接受。在許多種不同的生物辨識技術中，本論文將深入探討以掌形特徵為基礎的掌形辨識技術，並建立具有極佳辨識率的身份辨識系統。系統架構主要包含四個模組：影像擷取模組、影像前處理模組、特徵萃取與分類辨識模組。首先，利用數位相機擷取掌形影像，再經由前處理擷取出掌形中的手指部分，接著以結合(2D)2-PCA與(2D)2-LDA，對影像的行與列取共變異矩陣，找出類別內（within-class）較集中的訓練資料。接著以這些資料的平均作為類別代表資料，最後以(2D)2-PCA找出能將這些代表資料在類別間（between-class）分散的降維矩陣。在分類辨識模組方面，即可利用掌形影像降維後的特徵向量進行使用者註冊或辨識。
資料庫是使用實驗室自建的，總共有131個自願者的手掌影像，每人取六張影像，總共786張手掌影像對系統進行測試。本研究主要以結合(2D)2-PCA與(2D)2-LDA為特徵萃取方式，以等錯率（equal error rate）做為辨識標準，辨識率最高為97.71%，其中成功率最低為LDA特徵萃取方式，但仍有93%的辨識率。本論文會針對實驗結果進行分析比較，來驗證本系統所提的相關理論，以供後續研究作為參考。

關鍵字

生物辨識 ； 掌形辨識

並列摘要

In recent years, with an increasing emphasis on security, personal authentication based on biometrics has been receiving extensive attention. Among many different biometric technologies, this thesis examines hand-shape recognition technique for personal identification and develops a good performance recognition system based on human hand shape features. The proposed system includes four modules: image acquisition, image pre-processing, feature extraction, and recognition modules.First, the system captures a hand image using digital camera, then uses pre-processing algorithms to obtain the hand of fingers from the hand image via image pre-processing module. The feature extraction module adopts combination (2D)2-LDA and (2D)2-PCA method. It works simultaneously in the row and column directions of hand images.First, we find the more concentrated training samples, and calculate the sample mean in each class. Second, we employ the (2D)2-PCA to find projection matrix which can scatter the sample mean between classes. Eventually, the system applies these projected feature vectors for hand matching in recognition module.
Database of Hand images are from VIPCCL lab.Total volunteers are 131 people. We takes 6 images from each person,so the system tests 786 hand images.The main extraction method is combination (2D)2-LDA and (2D)2-PCA.Using equal error rate to be the recognition criterion.The highest recognition rate is 97.71%,and the lowest is LDA feature extraction method.But the recognition rate still over 93%. This thesis analyzes the experimented results and verifies the related inferences of the proposed system for providing useful information for further research.

並列關鍵字

(2D)2-PCA ； (2D)2-LDA ； Hand Shape Recognition ； Biometrics

參考文獻

[1] R. P. Miller, “Finger dimension comparison identification system,” US Patent, No.3576538, 1971.

[2] R. H. Emst, “Hand id system”, US Patent, No. 3576537, 1971.

[3] H. Jacoby, A. J. Giordano, and W. H. Fioretti, “Personnel identification apparatus,” US Patent, No. 3648240, 1986

[4] H. Yoshikawa and S. Ikebata, “A microcomputer-based personal identification system,” Proc. of International Conference on Industrial Electronics, Control and Instrumentation, vol. 1, pp. 105-109, 1984.

[5] J. Svigals, “Low cost personal identification verification device based on finger dimensions,” IBM Technical Disclosure Bulletin, vol. 25, No. 4, 1982.