1. WLD: A Robust Local Image Descriptor Jie Chen, Shiguang Shan, Chu He, Guoying Zhao, Matti Pietikäinen, Xilin Chen, Wen Gao 报告人：蒲薇榄

2. 作者简介 Jie Chen  University of Oulu  Senior researcher  Department of computer science and engineering  Center for Machine Vision Research  jie.chen@ee.oulu.fi jie.chen@ee.oulu.fi Biography  His research interests include pattern recognition, computer vision, machine learning, dynamic texture, and watermarking.  He has authored more than 20 papers in journals and conferences, and he is a member of the IEEE.

3. 作者简介 Shiguang Shan  Chinese Academy of Science  Professor  Institute of computing technology  Key Lab of Intelligent Information Processing(IIP)  sgshan@ict.ac.cn sgshan@ict.ac.cn Biography  His research interests cover image analysis, pattern recognition, and computer vision.  He is focusing especially on face recognition related research topics, and has published more than 100 papers on the related research topics.

4. 作者简介 Matti Pietikäinen  University of Oulu  Professor  Department of computer science and engineering  Center for Machine Vision Research  mkp@ee.oulu.fi mkp@ee.oulu.fi Biography  His research interests include texture-based computer vision, face analysis, human motion analysis, human-computer interaction, person identification, and visual surveillance.  He has authored more than 250 refereed papers in international journals, books, and conference proceedings, and about 100 other publications or reports. His publications have more than 1,800 citations in the Science Citation Index.

5. 文章出处 Publication  IEEE Transactions On Pattern Analysis and Machine Intelligence, vol. 32, no. 9, September 2010. References  T. Ahonen, A. Hadid and M. Pietikäinen "Face Description with Local Binary Patterns: Application to Face Recognition", IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 28, no. 12, pp.2037 -2041 2006  J. Anderson An Introduction to Neural Networks, 1995 :The MIT Press  J. Begard, N. Allezard and P. Sayd "Real-Time Human Detection in Urban Scenes: Local Descriptors and Classifiers Selection with AdaBoost-like Algorithms", Proc. IEEE Int\'l Conf. Computer Vision and Pattern Recognition Workshops, 2008  J. Chen, S. Shan, G. Zhao, X. Chen, W. Gao and M. Pietikäinen "A Robust Descriptor Based on Weber\'s Law", Proc. IEEE Int\'l Conf. Computer Vision and Pattern Recognition, 2008  ……

6. Abstract Inspired by Weber’s Law, this paper proposes a simple, yet very powerful and robust local descriptor, called the Weber Local Descriptor (WLD). It is based on the fact that human perception of a pattern depends not only on the change of a stimulus (such as sound, lighting) but also on the original intensity of the stimulus. WLD consists of two components: differential excitation and orientation. For a given image, we use the two components to construct a concatenated WLD histogram. Experimental results on texture classification and human face detection show a promising performance to other widely used descriptors.

7. 文章结构 Abstract Introduction WLD for Image Representation Application to Texture Classification Application to Face Detection More Experimental Validations and Discussions Conclusion

8. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

9. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

10. Descriptor Sparse Descriptor SIFT PCA-SIFT HOG RIFT Dense Descriptor LBP Gabor wavelet Cross Descriptor Use SIFT in dense way Use LBP in sparse way

11. SIFT Scale Invariant Feature Transform ，尺度不变特 征变换匹配算法，对旋转、尺度缩放、亮度变 化等保持不变性。 尺度空间极值检测。在尺度空间通过高斯微分函数 来检测潜在的对于尺度和旋转不变的兴趣点。 关键点定位。在兴趣点位置上，确定关键点的位置 和尺度。 方向确定。基于图像局部的梯度方向，给每个关键 点分配方向。 关键点描述。在每个关键点的领域内测量图像局部 的梯度。最终用一个特征向量来表达。

12. LBP

13. Weber’s Law 韦伯定律是表明心理量和物理量之间关系的定 律。德国生理学家和心理学家韦伯 (1795 － 1878) 发现同一刺激差别量必须达到一定比例，才能 引起差别感觉。这一比例是个常数，用公式表 示： ΔI/I = k ΔI ：差别阈限 I ：标准刺激强度 k ：常数 / 韦伯分数

14. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

15. WLD Weber Local Descriptor, is a dense descriptor Employs the advantages of SIFT and LBP SIFT: compute the histogram using the gradient and its orientation LBP: computational efficiency and smaller support regions Two components of WLD Differential excitation ( 差分激励 ) Orientation ( 方向 )

16. WLD 计算过程 差分激励 梯度方向

17. Differential Excitation(ξ )

19. Using arctangent function

20. Orientation(θ)

21. T=8

22. WLD Histogram j=1,2,…N-1, N is the dimensionality of an image t=1,2,…T-1, T is the number of the dominant orientations The size of this 2D histogram is T*C, where C is the number of cells in each orientation, and C=M*S.

23. WLD Histogram WLD=T*C T=8 M=6, S=3 C=M*S=18

24. Weight for WLD Histogram

25. Characteristics of WLD WLD is based on a physiological law, it has powerful representation ability for textures. The detected edges match the subjective criterion elegantly since WLD depends on the perceived luminance difference. WLD is robust to noise appearing in a given image. WLD is robust to changes in illumination.

26. Characteristics of WLD

27. Multiscale Analysis WLD P, R P SizeLength = 2*R+1 One can also generalize them to a circular one.

28. Comparison

30. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

31. Texture Classification Brodatz KTH-TIPS2-a

32. Texture Classification Set M = 6, T = 8, S = 20 Weight each subhistogram H m using the weights in the Weight for WLD Histogram table. Use the K-nearest neighbor as classifier(K=3) Use the normalized histogram intersection as similarity measurement of two histograms:

33. Texture Classification

34. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

35. Face Detection

36. Use a P-neighborhood WLD operator (P = 8 and R = 1) Divide a face sample of size w*h into K overlapping blocks (K = 9) of size (w/2)*(h/2) pixels. Set M = 6, T = 4, S = 3 Train a SVM classifier

37. Face Detection Training data set Positive set Sj Negative set Sn Test data set MIT+CMU frontal face test set Aleix Martinez-Robert(AR) face database CMU profile testing set

38. Face Detection

40. MIT+CMU frontal face test set Aleix Martinez-Robert(AR) face database CMU profile testing set

41. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

42. WLD arctan V.S. WLD log

43. The Effect of Parameters

44. Different Filters

45. Components

46. Robustness to Noise

47. 讲解提纲 相关知识介绍 用于图像描述的 WLD 算子 WLD 应用于纹理分类 WLD 应用于人脸检测 WLD 深入探讨 结论

48. Conclusion This paper propose a novel discriminative descriptor called WLD. It is inspired by Weber’s Law, and it consists of two components: differential excitations and orientations. Experimental results show that WLD illustrates a favorable performance on both texture classification and face detection. Future interest lies in how to exploit the proposed descriptor for the domain of face recognition and object recognition