Presentation is loading. Please wait.

Presentation is loading. Please wait.

On the relevance of facial expressions for biometric recognition Marcos Faundez-Zanuy, Joan Fabregas Escola Universitària Politècnica de Mataró (Barcelona.

Similar presentations


Presentation on theme: "On the relevance of facial expressions for biometric recognition Marcos Faundez-Zanuy, Joan Fabregas Escola Universitària Politècnica de Mataró (Barcelona."— Presentation transcript:

1 On the relevance of facial expressions for biometric recognition Marcos Faundez-Zanuy, Joan Fabregas Escola Universitària Politècnica de Mataró (Barcelona UPC) SPAIN

2 JCEE’07 15 Novembre 20072 OUTLINE Biometrics vs. facial analysis Transform methods for face recognition Experimental results –Identification –Verification Conclusion

3 JCEE’07 15 Novembre 20073 Biometrics vs. facial analysis (1/2) Biometric problem goal: a feature extraction insensitive to expression and highly discriminative among individuals. Expression analysis goal: a feature extraction insensitive to different individual’s variation and highly discriminative among expressions.

4 JCEE’07 15 Novembre 20074 Biometrics vs. facial analysis (2/2) Text-independent Speaker recognition goal: a feature extraction insensitive to the message content and highly discriminative among individuals. Speech recognition goal: a feature extraction insensitive to different individual’s variation and highly discriminative among phonemes. Same feature extraction works well for both!: MELCEPSTRUM

5 JCEE’07 15 Novembre 20075 Goal of this paper In this paper we use the Japanese Female Facial Expression Database (JAFFE) in order to evaluate the influence of facial expression in biometric recognition rates. In our experiments we used a nearest neighbor classifier with different number of training samples, different error criteria, and several feature extractions

6 JCEE’07 15 Novembre 20076 Pattern recognition system image 100 WHT/DCT coefficients Distance measure

7 JCEE’07 15 Novembre 20077 The Face Recognition Approaches Identification (1:N) –Pin-less access –It does not work for large populations Verification (1:1)

8 JCEE’07 15 Novembre 20078 The relevance of feature extraction It achieves a reduction on the number of data that must be processed, model sizes, etc., with the consequent reduction on computational burden. The transformation of the original data into a new feature space can let an easier discrimination between classes (faces).

9 JCEE’07 15 Novembre 20079 Feature extraction based on DCT DCT: It is a fast transform that requires real operations and it is a near optimal substitute for the KL transform of highly correlated images. It has excellent energy compactation for images. Applications: MPEG, JPEG, etc. DCT Dim. Reduction (10) + IDCT

10 JCEE’07 15 Novembre 200710 Feature extraction based on WHT The two- dimensional Hadamard transform pair for an image U of pixels is obtained by the equation:

11 JCEE’07 15 Novembre 200711 Computational burden of KLT, DCT and WHT for images of size N×N

12 JCEE’07 15 Novembre 200712 JAFFE DATABASE

13 JCEE’07 15 Novembre 200713 Experiment conditions

14 JCEE’07 15 Novembre 200714 Experimental results (1/8)

15 JCEE’07 15 Novembre 200715 Experimental results (2/8)

16 JCEE’07 15 Novembre 200716 Experimental results (3/8)

17 JCEE’07 15 Novembre 200717 Experimental results (4/8)

18 JCEE’07 15 Novembre 200718 Experimental results (5/8)

19 JCEE’07 15 Novembre 200719 Experimental results (6/8)

20 JCEE’07 15 Novembre 200720 Experimental results (7/8)

21 JCEE’07 15 Novembre 200721 Experimental results (8/8)

22 JCEE’07 15 Novembre 200722 Experimental results: summary Identification rates DCT

23 JCEE’07 15 Novembre 200723 Experimental results: summary Verification errors DCT

24 JCEE’07 15 Novembre 200724 Experimental results: summary Identification rates WHT

25 JCEE’07 15 Novembre 200725 Experimental results: summary Verification errors WHT

26 JCEE’07 15 Novembre 200726 CONCLUSIONS We have studied the relevance of facial expressions on biometric systems using two different feature extraction algorithms (DCT and WHT) and two different error criterion (MAD and MSE). Main conclusions are the following: –facial expression produces a drop on recognition rates (verification and identification applications) –Optimal vector length for biometric recognition seems to be same value than in the absence of facial expression (100 components). This value is quite stable for all the studied scenarios (transform, error criterion and different facial expressions).


Download ppt "On the relevance of facial expressions for biometric recognition Marcos Faundez-Zanuy, Joan Fabregas Escola Universitària Politècnica de Mataró (Barcelona."

Similar presentations


Ads by Google