近年來，生物特徵驗證引起了極大的關注，並在個人認證中發揮了重要的作用，而且已經有許多固有的生物特徵被廣泛的應用在個人認證系統，例如臉、指紋以及虹膜等。本篇論文著重在掌紋的認證。大部份傳統的掌紋認證方法都是使用接觸式設備來獲取掌紋影像，造成使用上的不方便和阻礙了應用的實用性。為了改進這些缺點，我們提出了一個非接觸的掌紋辨識方法，通過非接觸式採集設備獲取掌紋影像。 本論文首先對影像進行前處理來擷取感興趣的區域，然後使用尺度不變特徵轉換來提取影像的特徵。為了解決掌紋影像中的非線性變形，在特徵點的匹配之前，我們提出了一種自動分區的方法，在執行特徵點之前將掌紋影像分割成多個區域，每個區域的特徵點將分別各自匹配。最後對匹配的特徵點進行隨機抽樣一致和區域內部的優化，來刪除不符合拓樸關係的匹配點。將最終匹配的特徵點個數作為決策分數。實驗結果表明了所提出的方法在非接觸的掌紋認證中是有效果，且具可靠性。 ;Recently, biometric verification attracts tremendous attention and plays an important role in personal authentication. There are many inherent biometrics that have been widely used in personal authentication systems such as face, fingerprints and iris, etc. This thesis focuses on palmprint verification. Most of the traditional palmprint verification methods acquire palmprint images using contact acquisition devices, which is inconvenience in use and hinders the practicality in application. In order to improve these shortcomings, we propose a contactless palmprint identification method by acquiring palmprint images through contactless acquisition devices. In this thesis, the inputted image is first pre-processed to extract the region of interest (ROI), and then scale invariant feature transform (SIFT) is employed to extract the features of the image. To solve the non-linear deformation in the palmprint image, we propose an automatic partitioning method to divide the palmprint image into multiple regions before performing the matching of feature points. The feature points of each region will thereby be matched separately. Finally, the matched points are refined by employing random sample consensus (RANSAC) and optimization within the area to remove those matched points which fail to satisfy the topological relationships. The number of final matched SIFT points is then taken as the score for decision. Experimental results demonstrate that our proposed method is effective and robust in contactless palmprint verification.
