face recognition using the weber local descriptor

34
Face Recognition Using the Weber Local Descriptor 作作Dayi Gong Shutao Li Yin Xiang 作作作 作作作

Upload: ferdinand-joshua

Post on 02-Jan-2016

50 views

Category:

Documents


10 download

DESCRIPTION

Face Recognition Using the Weber Local Descriptor. 作者: Dayi Gong Shutao Li Yin Xiang 讲解 人: 余文倩. 作者简介. 龚大义 湖南大学 电气与信息工程学院 李树 涛研究生 [email protected]. 作者简介. 李树涛 湖南大学 教授,博士生导师 电气与信息工程学院 [email protected] 主要 研究 方向 : 图像处理 信息融合 压缩感知 稀疏表示 模式识别  机器学习. 作者简介. 向 荫 湖南大学 - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Face Recognition Using the Weber Local Descriptor

Face Recognition Using the Weber Local Descriptor

作者: Dayi Gong

Shutao Li

Yin Xiang

讲解人:余文倩

Page 2: Face Recognition Using the Weber Local Descriptor

作者简介

龚大义湖南大学电气与信息工程学院李树涛研究生[email protected]

Page 3: Face Recognition Using the Weber Local Descriptor

作者简介

李树涛湖南大学教授,博士生导师电气与信息工程学院[email protected]

主要研究方向 :图像处理 信息融合 压缩感知 稀疏表示 模式识别 机器学习

Page 4: Face Recognition Using the Weber Local Descriptor

作者简介

向荫湖南大学电气与信息工程学院李树涛研究生[email protected]

Page 5: Face Recognition Using the Weber Local Descriptor

文章出处

Publication Pattern Recognition (ACPR), 2011 First Asian Conference on Date 28-

28 Nov. 2011 , IEEE.

References J. Chen, S. Shan, C. He, et al. “WLD: a robust local image

descriptor,”IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 32, pp.1705–1720, September 2010

Page 6: Face Recognition Using the Weber Local Descriptor

Abstract

This paper presents a face recognition method using the Weber Local Descriptor (WLD).

The authors divide face images into a set of sub-regions and extract their WLD features respectively.

They introduce the Sobel descriptor to obtain the orientation component.

The experimental results over ORL and Yale face database verify the effectiveness of our method.

Page 7: Face Recognition Using the Weber Local Descriptor

文章结构

• Abstract

• Introduction

• The Extraction of WLD

• Face Recognition Method

• Experiments

• Conclusion

Page 8: Face Recognition Using the Weber Local Descriptor

讲解提纲

• 人脸识别

• WLD

• 基于 WLD 的人脸识别

• 实验结果分析

• 结论

Page 9: Face Recognition Using the Weber Local Descriptor

讲解提纲

• 人脸识别

• WLD

• 基于 WLD 的人脸识别

• 实验结果分析

• 结论

Page 10: Face Recognition Using the Weber Local Descriptor

人脸识别简介

• 人脸识别:特指利用分析比较人脸视觉特征信息进行身份鉴别的计算机技术。

Page 11: Face Recognition Using the Weber Local Descriptor

人脸识别流程

人脸识别

图像预处理

特征提取与选择

分类

直方图均衡化

中值滤波

基于子空间学习

贝叶斯

最近邻线性回归

基于几何特征

基于模板匹配

灰度拉升

LDAPCA

流行学习核方法

神经网络

Page 12: Face Recognition Using the Weber Local Descriptor

讲解提纲

• 人脸识别

• WLD

• 基于 WLD 的人脸识别

• 实验结果分析

• 结论

Page 13: Face Recognition Using the Weber Local Descriptor

A. Differential Excitation

The Extraction of WLD

B. OrientationSobel operator

C.WLD Histogram

Page 14: Face Recognition Using the Weber Local Descriptor

Sobel Operator

图像处理算子之一,主要用于边缘检测它是一种离散性差分算子,用来运算图像

亮度函数的梯度之近似值在图像的任何一点使用此算子,通过 3×3

模板作为核与图像中的每个像素点做卷积和运算,然后选取合适的阈值以提取边缘。

Page 15: Face Recognition Using the Weber Local Descriptor

传统的 Sobel 算子

A. 检测水平边缘 B. 检测垂直边缘-1 0 1

-2 0 2

-1 0 1

1 2 1

0 0 0

-1 -2 -1

A1 A2 A3

A4 (X,Y) A5

A6 A7 A8

Page 16: Face Recognition Using the Weber Local Descriptor

传统的 Sobel 算子

与图像做卷积分别得到横向与纵向的亮度差分近似值

图像的每一个像素的横向及纵向梯度近似值可以用一下的公式结合,来计算梯度的大小

𝐺 𝑋=[− 10+1− 20+2− 10+1]∗ 𝐴 ,𝐺𝑦=[+1+2+1

000−1+2+1 ]∗𝐴

𝐺=√𝐺𝑥2+𝐺𝑦

2

Page 17: Face Recognition Using the Weber Local Descriptor

Sobel Operator

Page 18: Face Recognition Using the Weber Local Descriptor

Sobel Operator

Page 19: Face Recognition Using the Weber Local Descriptor

Sobel Operator

Page 20: Face Recognition Using the Weber Local Descriptor

Why Sobel Operator

In the original methods of WLD, the gradient information is extracted by the two neighboring pixels in vertical direction, and another two in the horizontal direction of current pixel. It is easy disrupted by noises.

The convolution template of Sobel operator with different weights is used to suppress the noise.

So the Sobel operator is more appropriate to extract the gradient orientation.

Page 21: Face Recognition Using the Weber Local Descriptor

A. Differential excitation

The Extraction of WLD

B.OrientationSobel operator

C.WLD Histogram

Page 22: Face Recognition Using the Weber Local Descriptor

讲解提纲

• 人脸识别

• WLD

• 基于 WLD 的人脸识别

• 实验结果分析

• 结论

Page 23: Face Recognition Using the Weber Local Descriptor

A. Preprocessing with Gaussian filter —to make the face image smoother

Face Recognition Method

= 𝐺 (𝑥 , 𝑦 , 𝛿 )= 12𝜋 𝛿2 exp(−

𝑥2+𝑦 2

2𝛿2 )

Page 24: Face Recognition Using the Weber Local Descriptor

B. Feature extractionThe face images are divided into a set of sub-regions. Feature extraction is accomplished by obtaining the WLD histogram feature

of each sub-image.

Face Recognition Method

Page 25: Face Recognition Using the Weber Local Descriptor

C. Decision fusionTo improve the performance of the recognition scheme, all recognition

results of the sub-images are dealt with by decision fusion through voting as:

Face Recognition Method

𝑉=max𝑖𝐹 𝑖 (𝑣 𝑗 ) , 𝑗=0,1 ,…,𝑛

Page 26: Face Recognition Using the Weber Local Descriptor

Face Recognition Method

A CB

Page 27: Face Recognition Using the Weber Local Descriptor

讲解提纲

• 人脸识别

• WLD

• 基于 WLD 的人脸识别

• 实验结果分析

• 结论

Page 28: Face Recognition Using the Weber Local Descriptor

A. Experiments on effect of the different parameters

D=TxN

Experiments

Page 29: Face Recognition Using the Weber Local Descriptor

A. Experiments on effect of the different parameters

Divided into X*Y sub-regions

Experiments

Page 30: Face Recognition Using the Weber Local Descriptor

Experiments

B. The comparison of WLD,LBP and LTP

Page 31: Face Recognition Using the Weber Local Descriptor

Experiments

C. Comparison with different methods

Page 32: Face Recognition Using the Weber Local Descriptor

讲解提纲

• 人脸识别

• WLD

• 基于 WLD 的人脸识别

• 实验结果分析

• 结论

Page 33: Face Recognition Using the Weber Local Descriptor

Conclusion

In this paper, we have presented a new face recognition algorithm based on WLD, which makes a contribution to improve the recognition accuracy.

Experimental results show that the WLD feature has a powerful representation in face recognition, which is robust to variations in facial expression, illumination condition, pose, partial occlusions etc.

In the future, we will investigate to fuse WLD with other effective features to make further improvement in face recognition field.

Page 34: Face Recognition Using the Weber Local Descriptor

Thank you!