1. 12009/12/16 Hierarchical Ensemble of Global and Local Classifiers for Face Recognition Yu Su, Shiguang Shan, Member, IEEE, Xilin Chen, Member, IEEE, and Wen Gao, Fellow, IEEE

2. 22009/12/16 I.INTRODUCTION This paper proposes a novel face recognition method which exploits both global and local discriminative features. Global features are extracted by 2-D discrete Fourier transform. Local feature are extracted by Gabor wavelet transform. The combination of global and local features plays a key role. Ensemble is also a key contributor to improve generalizability

3. 32009/12/16 I.INTRODUCTION Different Roles of Global and Local Features

4. 42009/12/16 4 II. EXTRACTION 2-D Discrete Fourier Transformation ( DFT ) A 2-D image of size M by N pixels, 0 ≦ u ≦ M - 1 and 0 ≦ v ≦ N - 1 are frequency variables. R( u, v ) and I( u, v ) is real and imaginary components. Extraction of Global Fourier Features

5. 52009/12/16 5 II. EXTRACTION Global feature extraction by 2-D DFT Some examples of inverse transform by using only the low frequency bands (about 30% of all the energy). The real and imaginary components named global Fourier feature vector (GFFV). The real and imaginary components named global Fourier feature vector (GFFV).

6. 62009/12/16 6 II. EXTRACTION Extraction of Local Gabor Features Gabor Wavelet Transform ( GWT) Gabor wavelet consists of a planar sinusoid multiplied by a 2-D Gaussian.

7. 72009/12/16 7 II. EXTRACTION Gabor features are grouped into a number of feature vectors named local Gabor feature vector (LGFV)

8. 82009/12/16 II. EXTRACTION Patch Selection via Greedy Search in this paper, we propose a patch selection method to automatically determine the positions and sizes of the local patches

9. 92009/12/16 9 Two layers of ensemble: III. COMBINING Construction of Hierarchical Ensemble Classifier L i : Local Gabor feature vector (LGFV) C L i : Local Component Classifier (LCC) C L : Local Ensemble Classifier (LEC) C G : Global Classifier (GC) C H : Hierarchical Ensemble Classifier (HEC)

10. 102009/12/16 10 III. COMBINING Firstly, the face images can be divided into two classes named intrapersonal pairs and interpersonal pairs. Weight Learning for Component Classifiers Secondly, for each image pair, a similarity vector can be obtained. Last step, two classes of the N–dimensional samples are fed into FLD to get an optimal linear projection from N-D to 1-D.

11. 112009/12/16 IV. RESULT Different Roles of Global and Local Features Gabor feature are more sensitive to the detailed local variations.

12. 122009/12/16 IV. RESULT The performance improvement becomes trivial when the number of LCCs exceeds 30. The performance of LEC is much better than that of the individual LCC (especially on Experiment 4).

13. 132009/12/16 IV. RESULT Hierarchical Ensemble Classifier

14. 142009/12/16 Thanks