sgshan-COMA-JOC-Final.doc

Size: px

Start display at page:

Download "sgshan-COMA-JOC-Final.doc"

务胥
7 years ago
Views:

1 285 vol. 28 No. 5 1) 1),2) 2) 1) 1),2) 1) ( ) 2) ( ) : Fisherface Fisherface FERET CAS-PEAL-R1 TP391 Curse of Mis -alignment Problem in Face Recognition SHAN Shi-Guang 1), GAO Wen 1),2), CHANG Yi-Zheng 2), CAO Bo 1), CHEN Xi-Lin 1),2) 1) (ICT-ISVISION JDL for Face Recognition, ICT, Chinese Academy of Sciences) 2) (School of Computer, Harbin Institute of Technology) Abstract: In this paper, we investigate the rarely concerned curse of mis -alignment problem in face recognition, and propose a novel perturbation learning solution. Mis -alignment problem is firstly empirically investigated through evaluating systematically the Fisherface s sensitivity to mis-alignment on the FERET face database by perturbing the eye coordinates, which reveals that the imprecise localization of the facial landmarks abruptly degrades the Fisherface system. We explicitly define this problem as curse of mis-alignment for highlighting its significance. Aiming at this problem, we propose a set of measurement combining the recognition rate with the alignment error distribution to evaluate the overall performance of specific face recognition approach with its robustness against the mis -alignment considered. Finally, a perturbation learning method, named E-Fisherface, is proposed to reinforce the recognizer to model the mis-alignment variations in the training stage. Experimental results on FERET and CAS-PEAL-R1 have impressively indicated the effectiveness of the proposed E -Fisherface to tackle the curse of mis -alignment problem. Keywords : Face Recognition, Feature Alignment, Linear Discriminant Analysis, Curse of Mis-alignment, Perturbation Learning ( ) 863 (2002AA118010)

2 285 vol. 28 No. 5 [1,2,3] 92x Tangent Distance [4] 5 (a) (b) (c) 1 a b c E-Fisherface Fisherface FERET 6 Fisherface [5] Fisherface [6]

3 285 vol. 28 No. 5 Fisherface Belhumeur[5] PCA LDA 1000 FERET DARPA FERET [2] FERET FB 1 FERET [2] Fisherface Ltraining set Ttesting set G(gallery) P(probe set)[2] Fisherface PCA LDA W pca W lda L G P LDA P LDA G 1 FERET L Training set G P GGallery P T FB (Probes) 2 FERET FERET 2 FB Gallery FERET 3 3(a) htb x64 D 28 (Mask) 64x64 hd bd td D bd td hd (a) (b) (c) (d) (e) 3 Probe

4 285 vol. 28 No. 5 3D T x, T y s θ 4 (Appearance) 4 Fisherface FERET-FB 5 (a) (b) (c) 5 FERET -FB Fisherface (d) 5(a) 5(b) 5(c) 5(d) Fisherface

5 285 vol. 28 No. 5 6(a-c) 6(d-f) def Fisherface Fisherface (a) (b) (c) (d) (e) (f) 6 [4] OCR 1 / 2 compact Gabor [7,8,9,10,11] [8, 9] Gabor Lampinen[12]

6 285 vol. 28 No. 5 Gabor Gaussian Derivatives Gabor [13] High-Order Autocorrelation HOA [14] 3 [15, 16, 17] Martinez (Perturbation)[16] Gallery 4 Tangent Distance [4, 18] OCR (Robust Correlation)[19] (Global Affine Transform, GAT)[20] A B 7 A 92%B 100%C A 10 B A [21] 7 A,B,C ABC P* P P P* = d( P, P*) p( ) ~ N( µ, σ 2 )

7 285 vol. 28 No. 5 * 1 r Ω * r = P( ) r( ) d (1) Ω P( ) r( ) r * * r 7 A C C A 10 C 10 / 2 R r( ) r R = P( ) d = r0 r Ω r 0 * 0. (2) R 01 1 R 3 S S = * = 1/ R r0 / r. (3) S (1, ) 1 7 A B 0 1 p( ) ~ N(0 1, ) % 1~ 3 AB A B A B 7 R S 7 A C RS

8 285 vol. 28 No. 5 2 ABC r 0 (%) r*(%) R S A B C E-Fisherface Fisherface E-Fisherface LDA / Fisherface 1PCA Fisherface PCA PCA W pca 2 LDA 8 8 LDA PCA W pca LDA W lda 3 W pca W lda Fisherface Gallery LDA LDA Gallery LDA 8 8 E-Fisherface Martinez (Perturbation) [16] Martinez Gallery Gallery

9 285 vol. 28 No. 5 9 FERET-FB CAS-PEAL-R1 FERET FB E-Fisherface Fisherface 3 / Fisherface E-Fisherface 3 E-Fisherface Fisherface E-Fisherface (a) (b) (c) (d)

10 285 vol. 28 No E-Fisherface Fisherface 3 / r 0 (%) r*(%) R S Fisherface X E-Fisherface Fisherface Y E-Fisherface Fisherface E-Fisherface Fisherface E-Fisherface E-Fisherface r 0 CAS-PEAL 1,040 30,900 1,040 9,060 [22] CAS-PEAL-R1 [23][24] CAS-PEAL-R1 4 CAS-PEAL-R1 Fisherface E-Fisherface Fisherface E-Fisherface E-Fisherface

11 285 vol. 28 No. 5 Fisherface 5.3 Fisherface Fisherface E-FisherfaceFERET 1. Fisherface Fisherface Fisherface E-Fisherface Fisherface Gabor [1] Chellappa R., Wilson C.L., Sirohey S., Human and machine recognition of faces: a survey, Proceedings of the IEEE, 1995, 83(5): [2] Phillips P.J., Moon H., etc. The FERET evaluation methodology for face -recognition algorithms, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(10): [3] Philips P.J., Grother P.J., Micheals R.J., Blackburn D.M., E.Tabassi, and J.M.Bone, Face Recognition Vendor Test 2002: evaluation report, Technical Report, NISTIR 6965, National Institute of Standards and Technology, 2003 ( [4] Simard P.Y., LeCun Y.A., Denker J.S., and Victorri B, Transformation invariance in pattern recognition--tangent distance and tangent propagation. International Journal of Imaging System and Technology, 2001, 11(3): [5] P.N.Belhumeur, J.P.Hespanha etc. Eigenfaces vs Fisherfaces: rec ognition using class specific linear projection, 1997, IEEE Transactions on Pattern Analysis and Machine Intelligence, 20(7): [6] Shan S., Chang Y., Gao W., Cao B., Curse of mis-alignment in face recognition: problem and a novel mis -alignment learning solution, in Proceeding of The 6th International Conference on Face and Gesture Recognition, Jeju, Korea, 2004, [7] Buhman J., Lades M., and Malsburg C. von der, Size and distortion invariant object recognition by hierarchical graph matching, in Proc. of IEEE International Joint Conference on Neural Network, San Diego, CA, Jan. 1989, vol.2: [8] Lades M., Vorbruggen J.C., etc, Distortion invariant object recognition in the dynamic link architecture, IEEE Trans. On Computers, 1993, 42(3): [9] Wiskott L, Fellous J.M, Kruger N., Malsburg C. von der, Face recogniton by elastic bunch graph matching, IEEE Transactions on Pattern Analysis and Machine Intelligence, 1997, 19(7):

12 285 vol. 28 No. 5 [10] Rolf P. Wurtz, Object recognition robust under translations, deformations, and changes in background, IEEE Transactions on Pattern Analysis and Machine Intelligence, 1997, 19(7):pp [11] Shan S., Gao W., Chang Y., Cao B., Yang P., Review the strength of Gabor features for face recognition from the angle of its robustness to mis-alignment, in Proceeding of International Conference on Pattern Recognition, Cambridge, UK, 2004, vol.1, pp [12] Lampinen J. and Oja E.. Distortion tolerant pattern recognition based on self-organizing feature extraction. IEEE Transactions on Neural Networks, 1995, 6(3): [13] Schiele B, Crowley J. L., Recognition without correspondence using multidimensional receptive field histograms, International Journal of Computer Vision, 2000, 36(1):31-50 [14] Perantonis S. J. and Lisboa P. J. G., Translation, rotation, and scale invariant pattern recognition by high-order neural networks and moment classifiers, IEEE Trans. Neural Network, Mar vol.3: , [15] Thien M. Ha and Horst Bunke, Off-Line, Handwritten numeral recognition by perturbation method, IEEE Transactions On Pattern Analysis and Machine Intelligence, 1997, 19(5,): , [16] A.M. Martinez. Recognizing imprecisely localized, partially occluded and expression variant faces from a single sample per class, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24(6): [17] Shan S.G., Gao W., Zhao D.B., Face Identification Based On Face -Specific Subspace, International Journal of Image and System Technology, 2003, 13(1):23-32 [18] Keysers D., Macherey W., Ney H., Dahmen J., Adaptation in statistical pattern recognition using tangent vectors. In IEEE Transactions on Pattern Analysis and Machine Intelligence, 2004, 26(2): [19] Jonsson K., Kittler J., Li Y., Matas J.: Support vector machines for face authentication. Image and Vision Computing, 2002, 20(5-6): [20] Wakahara T., Kimura Y., Tomono A., Affine-invariant recognition of gray-scale characters using global affine transformation correlation, IEEE Trans. on Patten Analysis and Machine Intelligence, 2001, 23(4): [21] Shan S., Study on Some Key Issuses in Face Recognition [Ph.D dissertation], Institute of Computing Technology, Chinese Academy of Sciences, Beijing, 2004 (In Chinese) [ ], 2004 [22] Cao B., Shan S., Zhang X., Gao W., Baseline Evaluations On The CAS-PEAL-R1 Face Database, in Lecture Notes in Computer Science (LNCS3338), Advances in Biometric Person Authentication, Guangzhou, 2004, [23] Viola P, Jones M., Rapid Object Detection using a Boosted Cascade of Simple Features, in Proc. of Internationa l Conference on Computer Vision and Pattern Recognition, Kawaii, 2001, vol. I: [24] Cao B., Shan S., Gao W., Zhao D., "Localizing the Iris Center by Region Growing Search," Proceedings of IEEE International Conference on Multimedia and Expo, Vol. 2, pp , ( ) 863 (2002AA118010)

13 285 vol. 28 No. 5 SHAN Shiguang, male, born in March 1975, Ph.D., Assistant Researcher in Institute of Computing Technology, Chinese Academy of Sciences. His research interests include pattern analysis, machine learning, and computer vision. He is especially absorbed in face recognition related research topics. GAO Wen, male, born in 1956, Ph.D., Professor in Institute of Computing Technology and Graduated School of Chinese Academy of Sciences. He is also a professor in Harbin Institute of Technology. His research interests include multimedia data compression, image processing, computer vision, multimodal interface, artificial intelligence and virtual reality. CHANG Yizheng, female, born in 1979, Master candidate in School of Computer, Harbin Institute of Technology. Her research is focused on face recognition. CAO Bo, male, born in 1977, Ph.D Candidate in Institute of Computing Technology, Chinese Academy of Sciences. He is currently working on face recognition. CHEN Xilin, male, born in 1965, Ph.D., Professor in Institute of Computing Technology, Chinese Academy of Sciences and School of Computer, Harbin Institute of Technology. His research interests include computer vision, pattern recognition, machine learning, and intelligent human-computer interface sgshan@ict.ac.cn Background of the article: This work is conducted in the ICT-ISVISION Joint R&D Lab for face recognition (hereinafter named by FRJDL), Institute of Computing Technology Chinese Academy of Sciences, which has been working on face recognition related researches for more than 9 years as the first specialized lab on face recognition in China. FRJDL has always been aiming at inventing fully innovative kernel technologies and dealing with key issues in Biometrics, especially in face recognition area, based on which, multiple disciplines such as pattern recognition, machine learning, and computer vision, are theoretically studied. In the last 9 years, plenty of research work has been done on the basic theories in face recognition, as well as the practical engineering techniques. About 100 research papers had been published by the FRJDL members. And many of our research fruits have been successfully applied to practical applications such as bank surveillance, access control, and time-card systems. The work presented in this paper is just motivated by one of our practical fully automatic face recognition systems. In the system, we noticed that a large percent of false identification events are caused actually by the inaccurate eye localization results. To improve the robustness of the system to this kind of errors, we have conducted the work presented in this paper.

untitled

untitled 100084 E-mail: ahz@mail.tsinghua.edu.cn (Look Up Table, LUT) Adaboost 300 Adaboost TP391 Real Time Facial Expression Classification WANG Yubo, AI Haizhou, WU Bo, HUANG Chang Computer Science and Technology