隨著電子時代的來臨，許多的資料都數位化後儲存於電腦系統上處理。這使得電腦系統的安全性變的愈加重要，尤其是遇到機密或個人隱私的資料時。對於系統安全的第一道防線，通常就是對於使用者的身分確認。而最常見的身分確認方式就是要求使用者輸入帳號與密碼。這種方法雖然簡單、方便，但是相對的危險性也高。假如帳號和密碼被複製或遭竊，就有可能被人侵入系統裡盜取資料。因此，若能與具有難以複製或遭竊特性的生物特徵做結合，便可以提昇系統的安全性，降低被非法使用者登入的風險。 可用來作為個人身分確認的生物特徵主要可分為兩類。一是行為上的特徵，例如：簽名、打字習慣…等；另一種則是生理學上的特徵，例如：指紋、掌紋、掌型、視網膜…等。本論文將以輸入行為此項生物特徵為對象，針對打字時間及打字時手部動作的影像兩部分，做個人身分確認。進入電腦系統時輸入的帳號與密碼有可能遭到複製或盜取，但是輸入文字時的按鍵時間及手的動作、位置等動態資料卻不容易模仿及複製。因此，將輸入帳號過程中每個按鍵的按下及放開等時間作為特徵，利用倒傳遞神經網路做身分辨識。另外，以攝影機取得輸入時手部動作的影像並經由比對來確認輸入者的身分。在原本的帳號、密碼之外多一層具有生物特徵的辨認機制，以提昇識別系統的效能，有效防止非法的使用者登入系統來增加電腦系統的安全性。實驗結果驗證本論文所提出之身份辨認系統的可行性與有效性。 A lot of information will be digitalized and stored on computer system as the coming of electronic era. When the data was concerned with private information or commercial secrets, the security of computer system becomes more and more important. A general way to protect data against leak is to verify the identify of user. In general, it can be achieved by asking user to input a set of account and password. It is a simple and convenient way. However, it accompanies more risks. If the account and password of a user was grabbed or duplicated, data that stored in the computer system may be stolen. If we combine biometrics features that cannot be easily duplicated or stolen with account-password system, it can greatly improve the trustworthy and secure of identity verification system. There are various kinds of biometrics that can be utilized for personal identification. In this thesis, the identity is identified of computer users by using typing biometrics. It is difficult to observe typing biometrics by naked eyes, but can be recorded by electronic devices. The identification system verifies keystroke dynamics of user’s login string as the characteristic patterns using neural network techniques both interkey and hold times were used as features for identifying computer users. The typing images that were captured by camera when users keyin their account are also adopted in our work. By combining keystroke dynamics with typing images, the performance of authentication system can be improved. Moreover, the developed system can prevent the occurring of forgery logins to certain extend. Experimental results verify the feasibility and validity of the proposed system in personal identification.
