結合影像與音訊之身份認證系統; Identity Authentication System Using Audio-visual Information

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/9654

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/9654

題名:	結合影像與音訊之身份認證系統;Identity Authentication System Using Audio-visual Information
作者:	莊函潔;Han-Cheih Chuang
貢獻者:	資訊工程研究所
關鍵詞:	生物認證;Biometrics Recognition
日期:	2008-07-18
上傳時間:	2009-09-22 11:52:48 (UTC+8)
出版者:	國立中央大學圖書館
摘要:	現今社會中，人類比以前更加在乎人身安全及隱私問題。隨著日益進步的科技發展，硬碟和攝影器材價格的下降，數位影音資訊變得更容易取得，自動化的身份辨識系統變成人類重要的訴求。利用人類生物特徵如人臉、指紋、聲紋、簽名…等做為身份認證的依據，因具有唯一性和便利性，所以成為最近很熱門的主題。近來有許多雙模組身份認證方法被提出，利用兩種以上的生物特徵進行辨識，以增加認證系統之可靠度。本論文提出一個基於視訊之身份認證系統，結合音訊及人臉影像進行身份之認證。本論文，藉由讓語者回答隨機的字彙，以避免發生事先錄音之情況，並使用語音辨識來進行通關密碼的驗證。而在人臉特徵抽取的部份，同時使用了靜態和動態特徵。因為靜態資訊較容易以圖片偽造，而語者說話時臉部某些特徵之變化卻是很難模仿的，因此除了人臉靜態資訊外，本論文還同時參考一些動態資訊，如唇型變化和嘴巴開合等不易偽造的特徵，以辨識語者身份。整個系統中，首先利用音訊資料進行關鍵字認證，辨識語者是否說出正確的關鍵字，並利用嘴唇資訊輔助辨認，當確認關鍵字正確後，則繼續進行語者身份辨識。身份辨識部份，抽取人臉五官靜態特徵和嘴唇、下巴範圍的移動變化等動態特徵，分別進行身份辨識。最後結合靜態和動態特徵的辨識結果進行身份判定，並進行身份之認證。本研究方法，在關鍵字辨識的正確率上達92.96%，而身份辨識正確率為98.89%。由實驗結果證實本論文所提出之系統具有很好之效果。 In modernized society, security and privacy are the two issues that people concern most. Due to the rapid progress of technology development, the price of hard discs and photographic equipments reduce dramatically such that digital video information can be accessed more easily. Therefore, there is a growing need for automatic identity recognition systems to protect the misusing of personal private information. Biometrics recognition by using biometrics features, such as facial features, finger-print, voice, and signature, is an ultimate technique to resolve the problem because of its uniqueness and convenience. There are many bimodal identity authentication systems proposed recently which use two and or more biometrics features to improve the reliability of the authentication system. In this thesis, we propose a video-based identity authentication system which uses both audio and visual information. By asking a user a random question, the system can operate without the need of pre-recorded voice patterns. Since static features can be usually deceived by using photos, the system uses both static and dynamic features extracted from face images because the variations of dynamic features of a speaker are hard to be forged. In our work, the lip movement and mouth opening patterns are simultaneously adopted as the dynamic features to recognize the speaker's identity. First, the audio information and the lip information are inputted to check if the user has said the correct password. After the password has been checked, the identity of the user is then to be verified. Here, the static features like facial features and the dynamic features like the movement of lips and the jaw are utilized in identity recognition.。Finally, the two recognition results are fused to determine the identity. The proposed system can achieve 92.96% accuracy in password checking, and 98.89% accuracy in face recognition. Experimental results verify the validity of the proposed system in identity authentication.
顯示於類別:	[資訊工程研究所] 博碩士論文

文件中的檔案:

檔案	大小	格式	瀏覽次數

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....