Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

28
Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns George D. C. Cavalcanti Tsang Ing Ren, Josivan R. Reis 2012 IEEE International Conference on Systems, Man, and Cybernetics October 14-17, 2012, COEX, Seoul, Korea

Transcript of Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

Page 1: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

George D. C. Cavalcanti

Tsang Ing Ren,

Josivan R. Reis

2012 IEEE International Conference on Systems, Man, and Cybernetics

October 14-17, 2012, COEX, Seoul, Korea

Page 2: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

2

Outline

Introduction Occlusion Detection Recognition Experiments and results Conclusion

Page 3: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

3

INTRODUCTION

Page 4: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

4

Introduction

ه Challenges of face recognition systems is the problem of occlusion.

ه Uncontrolled environments such as drastic change of lighting, change of expression, beards and occlusions.

Page 5: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

5

Proposed approach

ه Apply in the problem for face recognition with sunglasses and scarf occlusion.

ه Consider illumination, rotation and inclination problems.

Page 6: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

6

Flowchart

Occlusion

Non-occlusion

Page 7: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

7

OCCLUSION DETECTION

Page 8: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

8

ه The image is divided into equal parts that are classified into occluded and non-occluded using MultiLayer Perceptron (MLP)

Page 9: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

9

ه Occlusion detection is a classification problem

N: a set of training imagesX: input layerY: output layer

Indicate : 1: non-occluded-1: occlude

Partial Face Classifier Using LDA and MLP”. In Proceedings of the 2010. IEEE/ACIS 9th International Conference on Computer and Information Science (ICIS '10)

Page 10: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

10

RECOGNITION

Page 11: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

11

ه For the recognition, Local binary

Pattern(LBP) feature is used on the non-occluded image part.

ه LBP widely used in face recognitionه Discriminative powerه Computational simplicityه Robustness to changes in grayscale.

Page 12: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

12

But:

LBP is not efficient for drastic lighting variations.

Solve:↓

we use a Gradientface as a preprocessing step before LBP.

Page 13: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

13

GradientFace

ه It is insensitive to variations in illumination and stands out in face recognition applications.

ه Gradientface Method:1. Transforms to the gradient domain

2. Eliminate noise or shadow (Gaussian filter)

Page 14: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

14

ه Smooth the image through a convolution with a Gaussian function∗ is the convolution operator

σ is the Gaussian function

ه Compute the image gradient I convolving in directions x, y

ه Generate as result ,Gradientface

Page 15: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

15

Page 16: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

16

Local binary patterns

ln : Corresponds to the central pixel value lc : The 8-neigbor pixels values

Page 17: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

17

Recognition

Page 18: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

18

ه [Gradientface + LBP ] computational simplicity

robust to scales changes and illumination variations.

ه Define the similarity between the LBP histograms of each image a similarity distance is used [7].

[E. P. Xing, A. Y. Ng, M. I. Jordan, and S. Russell, “Distance metric learning with application to clustering with side information,” in Advances in Neural Information Processing Systems 15 , S. Becker, S.Thrun, and K. Obermayer, Eds. Cambridge, MA: MIT Press, 2003, pp. 505–512 ]

Page 19: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

19

EXPERIMENTS AND RESULTS

Page 20: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

20

Database

ه AR Faceه More than 4000 color images (70 men 56 women)

ه With different facial expressions, lighting conditions and occlusions (sun glasses and scarf)

Page 21: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

21

ه ORLه 40 subjects, 10 different images for each subject.ه Facial expressions (open or closed eyes)ه Facial details (glasses and without glasses)

Page 22: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

22

Experiments1- MLPClassifier

ه Occlusion detection

Page 23: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

23

Experiments2- Recognition

23

Database: AR Face

Page 24: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

24

Experiments2- Recognition

Database: ORL Face

Page 25: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

25

CONCLUSION

Page 26: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

26

Conclusion

ه Address the problem of face recognition with occlusion caused by sunglasses and scarf.

ه The Gradientface applied to image with illumination problem and used to pre-processing the image, improved the recognition.

ه Combination of pre-processing techniques and classifies can still demonstrate improvements in face recognition problems.

Page 27: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

27

Q&A

Page 28: Recognition of Partially Occluded Face Using Gradientface and Local Binary Patterns

END

Thanks