| 摘要: | 在科技越來越進步的這個時代，已經不再設限於使用滑鼠或鍵盤來控制儀器，隨著近幾年人機互動越來越興起，現在可在公共區域或是家中利用空中手寫或是手勢來控制儀器，而人機互動更是已經廣泛應用在遊戲、娛樂、健康照護、圖形識別、身分認證等等領域之中。
傳統在身分認證，通常是以輸入密碼或攜帶加密金鑰等等方式作為身分認證的依據，但此方式不僅隨時要記得密碼或隨時攜帶金鑰之外，也容易遭他人竊取複製並盜用；相反地，利用空中手寫簽名取得特徵，不但解決了上述的問題，也比起透過人臉、聲紋、指紋、虹膜等等傳統生物認證系統，必須依靠昂貴儀器來取得特徵，更能節省成本。
本論文使用Leap Motion體感裝置來取得使用者的簽名，並尋找簽名軌跡中轉折點的位置，作為筆劃切割的依據，再根據累積筆劃計算Shape Context、轉折點速度及曲率當作特徵，在特徵比對時使用動態時間校正(Dynamic Time Warping，DTW)計算兩筆簽名軌跡之間的距離。在實驗的部分，本論文針對不同閥值和不同比對距離的組合、是否有事先去除偏差過大的資料、資料分群等各種情況下所得到的實驗結果，用以分析本論文所提出的方法，應用在身分認證系統上的效能。由實驗結果顯示，此方法應用在身份認證系統上，確實有很好的效能。
;Human-computer interaction has been widely used in the areas of games, health-care, entertainment, pattern recognition, and identity authentication. Due to the booming emerging of technological progress, the limitation of using mouse or keyboard to control 3C device is no longer a must. With the fast development of human-computer interaction recently, it is now possible to use air-handwriting, gesture, and so on to control 3C devices in public areas or at homes.
Traditional identity authentication usually relies heavily on the using of passwords or cryptographic keys. However, it is not only needing to remember the password or keeping the cryptographic key all the time but also easily losing passwords or keys and thereby suffering from password theft. On the contrary, using the unique features of air-handwriting signatures for identity authentication can resolve these problems. Furthermore, it is also cheaper than other biological authentication systems which must rely on expensive devices to acquire features such as face, voiceprint, fingerprints, iris and so forth.
In this thesis, we use Leap Motion to obtain the signatures of users and find the location of turning points in the signature trajectory as the basis for stroke cutting. Then, calculate the Shape Context of accumulate strokes, velocity and curvature of turning points as the features. In features matching, Dynamic Time Warping (DTW) is employed to calculate the distance between two signatures. In our experiments, we try to figure out the experimental results in each case such as combinations of different thresholds and different matching distances no matter whether performing data analysis or data clustering in advance or not to analyze the performance of our proposed method. Experimental results demonstrate the excel performance of our proposed system in identity authentication. |
