人臉三維取像與辨識;Study of 3D Human Face Imaging and Identification

NCU Institutional Repository > 理學院 > 光電科學研究所 > 博碩士論文 > Item 987654321/71597

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/71597

題名:	人臉三維取像與辨識;Study of 3D Human Face Imaging and Identification
作者:	邱俐雯;Chiu,Li-Wen
貢獻者:	光電科學與工程學系
關鍵詞:	人臉辨識;條紋投影輪廓儀;傅立業輪廓術;重建;三維形貌;face recognition;Projected Fringe Profilometry;Fourier transform profilometry;reconstruct;three-dimensional shape
日期:	2016-08-29
上傳時間:	2016-10-13 13:19:36 (UTC+8)
出版者:	國立中央大學
摘要:	對於現今的量測技術，由於條紋投影輪廓儀 (Projected Fringe Profilometry，PFP) 具有非接觸式的特色，目前被廣泛使用於量測物體三維形貌，本論文即利用條紋投影輪廓儀中的傅立葉輪廓術(Fourier transform profilometry，FTP) 重建物體的三維形貌，其優點具有快速、精準度高且只需單一影像即可還原物體三維形貌。本實驗運用FTP於人臉輪廓上的還原，並在重建的人臉三維形貌上擷取特徵值，利用每個人特徵值具有差異性的特性，提出雙重辨識方法達到辨識的效果。但在實際拍攝時，人臉會因為傾斜與操作距離不同而產生計算之誤差，透過兩種傾斜修正方法將人臉正規化，並利用三角測距法推算出實際距離，使本系統具有小角度傾斜容忍度以及35~55公分的拍攝容許距離。我們將人臉資料庫中的85人相互比對，總共比對3570次，藉由雙重辨識方法辨識，辨識率高達99.91%。 ;Because of fringe projection profilometry (PFP) has the advantages of non-contact measuring method, and it is widely use for measuring the three-dimensional shape of objects now. In this study, using Fourier transform profilometry (FTP) to reconstruct the three-dimensional shape of an object, which has the advantages of fast, high accuracy and only a single image that can restore the three-dimensional shape of an object. So we use FTP to restore three-dimensional shape of human face and get the feature value from it. Because of each person has unique feature, we proposed the method for double-identification to identification. In the measuring process, there are some artificial errors which are occurred by human face tilting and shifting. To solve this problem, we proposed two kinds of tilt-correction methods to normalize each human face, and calculating actual distance by using optical triangulation. So that, it allows a small tilt angle tolerance for system and extend shooting distance to 35~55 centimeters. Finally, we compare 85 persons with each other from face database, a total of comparison is 3570 times. By using double-identification method to identify 85 persons for each other, the recognition rate reaches 99.91%.
顯示於類別:	[光電科學研究所] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	587	檢視/開啟

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.....