基於改良式三維灰階伴隨矩陣之虹膜辨識

隨著科技的日新月異，各個國家都利用各自所擁有的科技保護自己國家的公共安全。然而，在所有的技術之中，利用生物辨識技術來維護資訊安全已成為一種新的趨勢。在各種不同的辨識方法之中，以虹膜影像為基礎的辨識系統是一個極為穩定的系統。因此，本論文將深入地去探討以人眼虹膜影像為基礎的辨識技術。本論文人眼虹膜資料庫採用葡萄牙內貝拉大學計算機科學演算法暨影像分析學群（Soft Computing and Image Analysis Group, Department of Computer Science, University of Beira Interior）所提供的UBIRIS虹膜資料庫，它裡面包含了241類不同的人眼，每個人眼有五張影像，總共1,205張影像，分別對系統進行測試。當一張人眼影像進入本論文所提出的辨識系統之中，一開始先經由「影像前處理模組」進行一系列虹膜區域定位的動作，接著將進入「特徵萃取模組」進行虹膜影像的特徵萃取，在特徵萃取的方法上，本論文將以三種方法來測試我們所提出的虹膜辨識系統：分別是以二維的灰階伴隨矩陣、三維灰階伴隨矩陣和改良式三維灰階伴隨矩陣來進行虹膜影像的特徵萃取，進行完特徵萃取模組之後，系統將會取得足以代表使用者身份的虹膜特徵碼（Iris Code）。最後，在「分類辨識模組」我們採用支持向量機（Support Vector Machine）來進行特徵碼的註冊以及比對的動作。從最後的實驗數據看來，本論文使用原始的三維灰階伴隨矩陣來做特徵萃取，再以支持向量機進行分類辨識（Identification），最後的結果為71.0256%。在相同條件之下，若利用改良式三維灰階伴隨矩陣來萃取虹膜的特徵資訊，之後再使用支持向量機進行分類辨識，實驗結果可達99.7%的辨識率，改良後的演算法明顯地提升了原始的三維灰階伴隨矩陣的性能。

關鍵字

生物辨識；虹膜；灰階伴隨矩陣

並列摘要

Now the technological society is developing quickly, changing with each passing day, more and more countries have developed new technologies for homeland security. However, biometric technology becomes a new trend in information security. Among biometric technologies, the person’s identification system by using human iris images is one of the reliable solutions of information security. For the reason, we discuss the human iris image-based recognition technologies for high security applications. This thesis proposed an iris recognition system. It is implemented and tested on the UBIRIS database（Soft Computing and Image Analysis Group, Department of Computer Science, University of Beira Interior）which includes 1,205 images from 241 classes. Our proposed iris recognition system includes three main modules: image pre-processing module, feature extraction module, and pattern recognition module. The first step of iris recognition system module is image pre-processing module. It is performed segmentation to divide interesting region of iris from eye images. The second step of feature extraction module adopts 2D Gray Level Co-occurrence Matrix, original 3D Gray Level Co-occurrence Matrix, and modified 3D Gray Level Co-occurrence Matrix to extract discriminating texture features to generate iris feature codes. Finally, the system applies to personal identify by using support vector machine to classify feature codes. In this thesis, we use original three-dimensional gray level co-occurrence matrix to extract personal identify information form iris, and adopts support vector machine to identify person. The experimental results show 71.0256 % recognition rate. Under the same conditions of our proposed modified three-dimensional gray level co-occurrence matrix, the experimental results approach the recognition rate above 99.7 %. In this thesis, we improved gray level co-occurrence matrix in three-dimensional statistical pixels diversification. It can be used in the high security applications.