§ Browsing ETD Metadata
System No. U0002-0408201414492800
Title (in Chinese) 基於KINECT建立3D人臉模型用於人臉辨識系統
Title (in English) Building 3D face model used to face recognition base on Kinect
Other Title
Institution 淡江大學
Department (in Chinese) 電機工程學系碩士班
Department (in English) Department of Electrical and Computer Engineering
Other Division
Other Division Name
Other Department/Institution
Academic Year 102
Semester 2
PublicationYear 103
Author's name (in Chinese) 許維新
Author's name(in English) Wei-Hsin Hsu
Student ID 601440075
Degree 碩士
Language Traditional Chinese
Other Language
Date of Oral Defense 2014-07-09
Pagination 56page
Committee Member advisor - 謝景棠
co-chair - 蘇木春
co-chair - 謝君偉
Keyword (inChinese) 人臉辨識
Keyword (in English) Face Recognition
Other Keywords
Abstract (in Chinese)
Abstract (in English)
The applications of face recognition is increasing day by day. The studies of face recognition are mainly occlusions (sunglasses, hats, masks) , illumination (shadows, glare) , facial expressions, age variation , and pose changes (profile, pitch angle is too large) etc, and the side face recognition has been a difficult problem. We propose a simpler and faster method to create 3D face model. First, using ASM(Active Shape Model) to detection and get the color and depth image of face by Kinect, then we based on the information of the depth image portrayed face in opengl three-dimensional. This method retains much texture of information of the original face images, and to create a complete change of face uneven depth. It still has a good result of repairing the distortion in side face. We can get a set face images of the same person with different angles by the method proposed in this paper. In recognition part, we use PCA(Principal Component Analysis) to reduce the dimensions, and combined SVM(Support Vector Machine) to classify. Experiments show that the side face recognition can have good results.
Other Abstract
Table of Content (with Page Number)
誌謝	I
中文摘要	II
英文摘要	III
目錄	IV
圖目錄	VII
表目錄	IX
第一章 緒論	1
1.1 研究背景	1
1.2 研究動機	2
1.3 研究方法	2
1.4 論文架構	3
第二章 相關研究與技術	4
2.1 相關研究	4
2.1.1 人臉辨識	4
2.1.2 3D人臉建模	5
2.2 相關技術	8
2.2.1 YCbCr膚色偵測	8
2.2.2 數學形態學(Mathematical Morphology)	10 侵蝕	11 膨脹	12 斷開	13 閉合	14
2.2.3 直方圖等化	15
2.2.4 主成分分析(Principal component analysis)	17
2.2.5 支持向量機(Support Vector Machine)	19 線性支援向量機(Linear SVM)	20 非線性支援向量機(Nonlinear SVM)	22
2.2.6 主動形狀模型(Active Shape Model)	23 點分布模型(Point Distribution Model , PDM)	24 特徵點座標正規化	25 特徵點分布之主成分分析(PCA)	26 ASM模型校準(Aligning)	26
第三章 人臉辨識系統	28
3.1 大綱	28
3.2 訓練部分	29
3.2.1 Kinect彩色與深度視角修正	29
3.2.2 ASM擷取人臉影像	30
3.2.3 彩色影像前處理	31 YCbCr偵測膚色濾背景	31 形態學影像處理	32 RGB值前處理並轉成資料型態儲存	33
3.2.4 深度pixel長度比之校正	34
3.2.5 OpenGL創造3D人臉模型之方法	38
3.2.6 直方圖等化	40
3.3 辨識部分	41
第四章 實驗結果	42
4.1 實驗環境	42
4.2 實驗資料庫	42
4.3 實驗結果	45
4.3.1 3D人臉模型	45
4.3.2 人臉辨識	47 3D模型轉2D影像之辨識率	47 與一般影像進行訓練之人臉辨識比較	48 有無表情變化對辨識率之影響	49 所選取之特徵向量數對辨識率之影響	50 角度辨識	52
第五章 結論與未來展望	54
5.1 結論	54
5.2 未來展望	54
參考文獻	55
圖2.1 Candide人臉模型[5]	7
圖2.2 FaceGen Modeller人臉建模示意圖[6]	8
圖2.3 將原始影像轉成YCbCr影像	9
圖2.4 對YCbCr設閥值過濾膚色	10
圖2.5 對影像做侵蝕之結果	11
圖2.6 對影像做膨脹之結果	12
圖2.7 對影像做斷開之結果	13
圖2.8 對影像做閉合之結果	14
圖2.9 過暗、過亮及低對比度之影像及其像素累計直方圖	15
圖2.10 灰階影像經直方圖除理後之結果	16
圖2.11 彩色影像經直方圖除理後之結果	16
圖2.12 SVM超平面示意圖	19
圖2.13 超平面分類示意圖	20
圖2.14 本論文ASM選取之人臉特徵點	25
圖3.1 流程圖	28
圖3.2 Kinect彩色及深度鏡頭校正前後比較圖	29
圖3.3 ASM人臉偵測之結果	30
圖3.4 ASM所擷取之彩色及深度影像	30
圖3.5 利用YCbCr偵測膚色之結果	31
圖3.6 對YCbCr做3次斷開之結果	32
圖3.7 將人臉與背景區隔之結果	32
圖3.8 原始彩色影像	33
圖3.9 分離後之R、G、B影像	33
圖3.10相同物品和鏡頭之距離與影像中所占pixel數之關係圖	35
圖3.11 需校正深度pixel長度比之例子示意圖	36
圖3.12 因pixel間之距離造成旋轉3D人臉時產生縫隙	38
圖3.13以4點構成之正方形當成3D人臉模型之pixel	39
圖3.14 依本論文提出之方法所創立之3D人臉模型	40
圖3.15 直方圖等化之結果	40
圖3.16 各角度之平均臉	41
圖4.1 部分由3D模型轉成2D影像-無表情	43
圖4.2 部分由3D模型轉成2D影像-有表情	43
圖4.3 部分一般彩色影像-無表情	44
圖4.4 部分一般彩色影像-有表情	44
圖4.5 正臉	45
圖4.6 右側臉	45
圖4.7 左側臉	46
圖4.8 仰角臉	46
圖4.9 俯角臉	46

表4.1 3D模型轉2D影像之辨識率	47
表4.2 以一般影像進行訓練之辨識率	48
表4.3 有表情變化之辨識率	49
表4.4 無表情變化之辨識率	50
表4.5 取前5維特徵向量之辨識率	51
表4.6 取前15維特徵向量之辨識率	51
表4.7 取前20維特徵向量之辨識率	52
表4.8 平均角度辨識率	53
表4.9 各角度之辨識率	53
[1]Faruqe, Md Omar, and M. Al Mehedi Hasan. “Face recognition using PCA and SVM,” Anti-counterfeiting, Security, and Identification in Communication, 2009. ASID 2009. 3rd International Conference on. IEEE, 2009.
[2]Li, Billy YL, et al. “Using kinect for face recognition under varying poses, expressions, illumination and disguise,” Applications of Computer Vision (WACV), 2013 IEEE Workshop on. IEEE, 2013.
[3]何宗庭, 王順吉, “以單張正面2D人臉影像之特徵點為基礎重建3D人臉,” 國防大學理工學院資訊工程學系, 民國100年5月。
[4]Jorgen Ahlberg, “CANDIDE-3–AN UPDATED PARAMETERISED FACE,” Jan. 2001.
[5]CANDIDE, http://www.bk.isy.liu.se/candide/
[6]Facegen Modeller, http://www.facegen.com/modeller.htm/
[7]V.Vapnik, The Nature of Statistical Learning Theory, New York: Springer-Verlag, 2000.
[8]蕭博文, “SVM based counterfeit banknote recognition,” 台灣科技大學電機工程系, 民國98年6月。
[9]T.F. Cootes, and C. J. Taylor, “Active shape models—‘smart snakes’,” Proceed-ings British Machine Vision Conference, Springer Berlin, 1992, pp. 266-275.
[10]R.H. Davies, T.F. Cootes, C.Twining and C.J. Taylor, “An Information theoretic approach to statistical shape modelling,” Proc. British Machine Vision Conference, 2001, pp.3-11.
Terms of Use
Within Campus
I request to embargo my thesis/dissertation for 2 year(s) right after the date I submit my Authorization Approval Form.
Agree to authorize disclosure on campus
Duration for delaying release from 2 years.
Outside the Campus
I grant the authorization for the public to view/print my electronic full text with royalty fee and I donate the fee to my school library as a development fund.
Duration for delaying release from 2 years.

If you have any questions, please contact us!

Library: please call (02)2621-5656 ext. 2487 or email