SVM [6] PCA+SVM 79.75% 9 FERE FERE. PCA LDA Adaboost SVM 5 1 SVM Moghaddam [6] M (x,y ) x R N y x y {0,1} M f ( x) = y α k( x, x ) + b x k f(x) = 1 x

Size: px

Start display at page:

Download "SVM [6] PCA+SVM 79.75% 9 FERE FERE. PCA LDA Adaboost SVM 5 1 SVM Moghaddam [6] M (x,y ) x R N y x y {0,1} M f ( x) = y α k( x, x ) + b x k f(x) = 1 x"

归闾
5 years ago
Views:

1 PCA Fsher FLD Adaboost SVM 9 FERE Adaboost 1. Gollomb [1] SEXNE Cottrell [] BP Edelman [3] Alce [4] PCA PCA [5] Moghaddam [6] (SVM) RBF Fsher FLD FERE 3.4% Shakhnarovch [7] Adaboost 78% ±15 ±30 SVM PCA+SVM PCA+Adaboost PCA+Adaboost+SVM FLD+CSCIE Centralzed Sample s Classfcaton Informaton Extractng [8] +SVM 5 SVM ( ) 1

2 SVM [6] PCA+SVM 79.75% 9 FERE FERE. PCA LDA Adaboost SVM 5 1 SVM Moghaddam [6] M (x,y ) x R N y x y {0,1} M f ( x) = y α k( x, x ) + b x k f(x) = 1 x [9] PCA+SVM PCA SVM x R N Σ=E[(x-µ) (x-µ) ] µ Σ m U=(u 1,u,,u m ) x y=u (x-µ) y PCA 3 PCA+ADABOOS Boostng [10] Adaboost[11][1] Boostng Vola [6] Adaboost PCA u g( x) = u x + c

3 c Adaboost c Adaboost 1, = = g x threshold h x t t t ( ) ( ) α 1 0, otherwse [1] 4 PCA+ADABOOS+SVM Adaboost [8] Adaboost Adaboost SVM 5 FLD+CSCIE+SVM [8] [14] FLD [8] Fsher ϖ 1 ϖ S = E[( x µ ) ( x µ ) ϖ ] = 1, S w = P( ϖ 1 ) S1 + P( ϖ ) S ) P( ϖ S = µ µ )( µ µ ) + ( µ µ )( µ ) Fsher J b ( 1 1 µ w Sbw ( w) w R n w S w F = w n w J F (w) w S w = S w b λ w c w c 1 1. FLD 1 3

4 x [ α 1, α ] x (, α1 ) ( α, + ) H P = E[logp( x)] Fsher S j j ~ γ = λ S j j P γ j γ j w P λ γ j j j w γ j = u j S u j u j S w j S j S S1 S j u j w J ( u j ) = = 1 ~ γ log ~ γ j j. 9 d J u ) J( u 1 ) J( u d ) J( u n ) d ( j FLD CSCIE FLD FLD CSCIE d FLD 1 CSCIE 1 d m SVM FERE

5 5 9 ±5 o 1.05 1 500 130 9 9 0 FERE U.S.

5 5 5 9 ±5 o FERE U.S. Army Research Laboratory FERE [6] FERE * [15] 4 a 4 b 4 c w h w h 5

4. 3.3 5 SVM 9 FERE w h 40 1 1 11 1. SVM 13 PIII-600, 9 100.0% 100.0% 100.0% 1, 9 FERE 90.88% 83.59% 95.

6 SVM 9 FERE w h SVM 13 PIII-600, % 100.0% 100.0% 1, 9 FERE 90.88% 83.59% 95.04% 100% FERE 90% [6] FERE 9 9 FERE 9 PCA+SVM PCA+Adaboost PCA+Adaboost+SVM FLD+CSCIE+SVM 9 PCA+Adaboost m SVM 4 PCA Adaboost PCA+Adaboost 3 SVM FLD+CSCIE 6

7 5. PCA+SVM PCA+Adaboost PCA+Adaboost+SVM FLD+CSCIE+SVM 1 FLD 7 PCA+Adaboost PCA 3,4 Adaboost 6. 5 Adaboost PCA PCA PCA+SVM SVM [4] 7

3.4 SVM 40 40 SVM 9 FERE 16 16 7. PCA+SVM PCA+Adaboost 8 4 4 3 3 40 40 ( 16 16 4 4 ) ( 3 3 40 40 ) SVM SVM [6] 9 FERE 8. SVM. 9 FERE SVM 9 FERE Dm 16 16 16 16 4 4 3 3 40 40 4 4 3 3 40 40 04 04 15 739 870 801 14 487 08 08 08 741 884 814 145 501 % 96.

8 3.4 SVM SVM 9 FERE PCA+SVM PCA+Adaboost ( ) ( ) SVM SVM [6] 9 FERE 8. SVM. 9 FERE SVM 9 FERE Dm % % %

9 FERE BMP JPG Adaboost [13] Sobel [15] 3 3 PCA+SVM %

10 PCA ( ) [1] B. A. Golomb, D.. Lawrence, and. J. Sejnowsk. SEXNE: A neural network dentfes sex from human faces. In Advances n Neural Informaton Processng Systems, pp , [] G. W. Cottrell and J. Metcalfe. EMPAH: Face, emoton, and gender recognton usng holons. In Advances n Neural Informaton Processng Systems, pp , [3] B. Edelman, D. Valentn, H. Abd. Sex classfcaton of face areas: how well can a lnear neural network predct human performance. Journal of Bologcal System, Vol. 6(3), pp , [4] Alce J.O'oole et al. he Percepton of Face Gender: he Role of Stmulus Structure n Recognton and Classfcaton. Memory and Cognton, Vol. 6, pp , [5] Alce J.O'oole, homas Vetter, Henrch H.Bulthoff & Nkolaus F.roje. he role of shape and texture nformaton n sex classfcaton. echncal Report No.3 December 10, [6] B. Moghaddam and M.H. Yang.Gender Classfcaton wth Support Vector Machnes. IEEE ransactons on Pattern Analyss and Machne Intellgence, Vol. 4, No. 5, pp , May 00 [7] G. Shakhnarovch, P.. Vola and B. Moghaddam. A Unfed Learnng Framework for Real me Face Detecton and Classfcaton. IEEE conf. On AFG 00. [8] BIAN Zhaoj, ZHANG Xuegong et al., Pattern Recognton, nd verson, snghua Unversty Press, Jan [9] V. N. Vapnk, ( ) V. N. Vapnk, (translated by ZHANG Xuegong), he Nature of Statstcal Learnng heory, snghua Unversty Press, Sept [10] R. E. Schapre, he Boostng Approach to Machne Learnng An Overvew, MSRI Workshop on Nonlnear Estmaton and Classfcaton, 00. [11] Y. Freund and R. E. Schapre. A decson-theoretc generalzaton of on-lne learnng and an applcaton to boostng. In Computatonal Learnng heory: Eurocolt 95, pp Sprnger-Verlag,

11 [1] Y. Freund and R. E. Schapre. Experments wth a New Boostng Algorthm. In Proceedngs of the 13 th Internatonal Conference on Machne Learnng, pp , Morgan Kaufmann, [13] P.Vola and M.Jones. Robust Real-tme Object Detecton. Cambrdge Research Laboratory, echncal Report Seres, February 001. [14],,., 00, 6(). YANG Jan, YANG Jngyu, LIU Nngzhong. he heory and Algorthm on Statstcal Uncorrelated Optmal Dscrmnant Analyss. Journal of Nanjng Polytechnc Unversty, 00, 6(). [15] L ZHUANG, Hazhou AI, Guangyou XU, Vdeo Based Face Recognton by Support Vector Machnes, Internatonal Conference on Computer Vson, Pattern Recognton and Image Processng, Mar. 8-14, 00, Durham, North Carolna USA. Face Gender Classfcaton Bo WU, Hazhou AI, Xpan XIAO, Guangyou XU Computer Scence and echnology Department, snghua Unversty State Key Laboratory of Intellgent echnology and Systems ABSRAC Gender classfcaton s a pattern recognton problem of decdng the gender accordng to the face mage. hs paper systematcally nvestgates the performance of several feature extracton and classfcaton methods on the gender classfcaton problem among whch nclude prncpal component analyss (PCA), Fsher lnear dscrmnant analyss (FLD), statstcally ndependent lnear dscrmnant analyss, Adaboost Algorthm, and Support Vector Machne (SVM). Comparatve expermental results on our 9-pose face database, the FERE face database, and a Web pcture face database are gven. he results show that the gender nformaton of human face clusters wthn certan sub-space, therefore proper compresson before gender classfcaton can greatly reduce the tme consumed, wthout sacrfcng good performance. Fnally the face-mage-based gender classfer system that s ntegrated wth an automatc face detecton and feature extracton platform s ntroduced. Keywords Gender classfcaton, prncpal component analyss, Fsher lnear dscrmnant analyss, Adaboost, Support Vector Machne 11

12 E-mal: wb98@mals.tsnghua.edu.cn E-mal: ahz@mal.tsnghua.edu.cn ,

Microsoft Word - 系统建设1.doc

Microsoft Word - 系统建设1.doc 计算机系统应用 http://www.c-s-a.org.cn 20 年第 20 卷第 4 期一种高性能近红外光人脸检测和眼睛定位算法张昌明, 童卫青, 王燕群 ( 华东师范大学计算机科学技术系, 上海 20024) 摘要 : 提出了一种新型的近红外光人脸检测和眼睛