Multiple Similarities Based Kernel Subspace Learning for Image Classification

Size: px
Start display at page:

Download "Multiple Similarities Based Kernel Subspace Learning for Image Classification"

Transcription

1 Multiple Similarities Based Kernel Subspace Learning for Image Classification Wang Yan, Qingshan Liu, Hanqing Lu, and Songde Ma National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, P.O. Box 2728, Beijing , China {wyan, qsliu, luhq, Abstract. In this paper, we propose a new method for image classification, in which matrix based kernel features are designed to capture the multiple similarities between images in different low-level visual cues. Based on the property that dot product kernel can be regarded as a similarity measure, we apply kernel functions to different low-level visual features respectively to measure the similarities between two images, and obtain a kernel feature matrix for each image. In order to deal with the problems of over fitting and numerical computation, a revised version of Two-Dimensional PCA algorithm is developed to learn intrinsic subspace of matrix features for classification. Extensive experiments on the Corel database show the advantage of the proposed method. 1 Introduction With image data growing rapidly, how to efficiently manage and browse images is an urgent and challenging problem. Image classification is an important technique for image browsing and retrieval in a large database [1], [2]. As a typical pattern recognition problem, image classification has two key issues, i.e., feature extraction and classifier selection based on extracted features. Most previous studies concentrate on designing the classifier, and directly take low-level visual cues as features, such as color, shape and texture [1], [2], [3]. Different feature vectors are concatenated end to end and form the feature vector of the image. The similarity between two images is usually measured in the Euclidean space with these feature vectors. Zheng, et al, propose using local preserving projection (LPP) [4], [5] to capture the local manifold of the images, but LPP is a linear approach in nature. Practically, as for some complex image data, especially nature images, there exist complex nonlinear variations, which will degrade the performance of the classification methods. Kernel Principal Component Analysis (KPCA) is a good nonlinear analysis method, which is actually a nonlinear version of Principal Component Analysis (PCA). Its idea is to first map the input data into an implicit feature space by a nonlinear mapping, and then the data are analyzed in the implicit feature space [6]. KPCA has been widely used in practical data analysis [6], [7], [8]. The new method proposed in this paper is inspired by KPCA. P.J. Narayanan et al. (Eds.): ACCV 2006, LNCS 3852, pp , c Springer-Verlag Berlin Heidelberg 2006

2 Multiple Similarities Based Kernel Subspace Learning 245 We first give a new perception of KPCA. It can be regarded as having two independent steps: kernel feature extraction and PCA on the kernel features [9]. Kernel feature vector of an image can be calculated as follows: construct an vector with kernel dot products between the image and all the training images first, and then center the vector by subtracting its mean value. Based on this perception, a scheme of matrix based kernel features for image classification is proposed. Since in image classification and retrieval, images are often described by multiple visual cues, such as color, shape and texture, if kernel dot products between two images on different visual cues are computed respectively, we can get a dot product vector between two images, and the kernel feature vector of an image becomes a kernel feature matrix. From the view that the kernel dot product being a similarity measure [10], [11], the kernel feature matrix provides a strategy to measure the multiple similarities between the image and training images, which should be more precise for image classification. In order to deal with the problem of the evaluation of eigenvectors, a revised version of Two- Dimensional PCA (2DPCA) [12] is developed to learn the intrinsic subspace of image feature matrices. Extensive experiments on the Corel database show that the proposed method has an encouraging performance. The rest paper is organized as follows: a new perception of KPCA is given in Section 2, and we present the proposed image classification scheme in Section 3. The experiments are reported in Section 4, followed by the conclusions in Section 5. 2 Kernel Principal Component Analysis The idea of KPCA is first to map the input data {x i } N i=1 into an implicit feature space F by a nonlinear mapping φ, andthepcaisperformedinf to get the nonlinear principal components of the input data [6]. It is unnecessary to know the mapping φ explicitly, and we only need to calculate the dot product between implicit features vectors {φ(x i )} N i=1 with a kernel function that satisfies Mercer s theorem [6]. Gaussian kernel is used in this paper for its popularity in image classification and retrieval [13], [14], and its definition is as follows: ( k(x 1, x 2 )=(φ(x 1 ) φ(x 2 )) = exp γ x 1 x 2 2). (1) For the following analysis, we define some symbols first. X =[x 1, x 2,, x N ] is the training set. The matrix Φ(X) =[φ(x 1 ),φ(x 2 ),,φ(x N )] is the mapping of training set in implicit feature space F.TheGrammatrixK=[K 1, K 2,,K N ], where the column vector K i is composed of dot products between φ(x i )andall the training set in F, i.e., K i =(k(x i, x 1 ),k(x i, x 2 ),,k(x i, x N )). It can be seen that K is symmetrical. KPCA is equivalent to solving the problem of eigenvectors and eigenvalues of covariance matrix C of {φ(x i )} N i=1 [6]. C = 1 N 1 (Φ(X) Φ(X)1 N )(Φ(X) Φ(X)1 N ) T, (2)

3 246 W. Yan et al. where 1 N is an N Nmatrix with each entry equals 1/N. Let W = [w 1, w 2,, w E ] denotes the unitary eigenvector matrix of C, wherew 1, w 2,, w E are unitary eigenvectors corresponding to positive eigenvalues λ 1,λ 2,,λ E,we get 1 N 1 WT (Φ(X) Φ(X)1 N )(Φ(X) Φ(X)1 N ) T W = Λ, (3) where Λ = diag(λ 1,λ 2,,λ E ). Since any eigenvector w i with eigenvalue λ i must lie in the span of {φ(x i )} N i=1 [6], we have W =(Φ(X) Φ(X)1 N )A, (4) where A = [α 1,α 2,,α E ]isan E matrix called eigenvector expansion coefficient matrix, and w i =(Φ(X) Φ(X)1 N )α i. Combining (3) and (4), and because Φ T (X)Φ(X) =K, weget 1 N 1 AT ( K K1 N )( K K1 N ) T A=Λ, (5) where K = K 1 N K. In fact, each column of K equals the corresponding column of K subtracting its mean value, so we call K centered Gram matrix. Thus, the solution of equation (5) is equivalent to solving the eigenvectors and eigenvalues of the covariance of K. Finally we normalize α 1,α 2,,α E in order to make (w i w i ) = 1. Note that K K1 N is symmetrical and all of its eigenvalues are nonnegative, it can be proved that any eigenvector α i of ( K K1 N )( K K1 N ) T /(N 1) with eigenvalue λ i is eigenvector of ( K K1 N ) with eigenvalue (N 1)λi. So the normalization condition is as follows: 1=(w i w i )=((Φ(X) Φ(X)1 N )α i (Φ(X) Φ(X)1 N )α i ) = α T i ( K K1 N )α i, (6) = (N 1)λ i (α i α i )= λ i (α i α i ) where λ i is the eigenvalue of ( K K1 N )( K K1 N ) T. As for test samples T =[t 1, t 2,, t L ], their projections in KPCA subspace are Y = W T (Φ(T) Φ(X)1 N )=W T Φ(T) W T Φ(X)1 N, (7) where 1 N is the N L matrix with each entry equals 1/N. Since all columns of W T Φ(X)1 N are identical, it is irrelevant for classification problem when using Euclidean distance. Combining (4) and (7) we get Y = W T Φ(T) =A T (Φ(X) Φ(X)1 N ) T Φ(T). (8) Define matrix K test =(k(x i, t j )) N L for test points, and then we get where K test = K test 1 N K test. Y =A T (K test 1 N K test )=A T K test, (9)

4 Multiple Similarities Based Kernel Subspace Learning 247 From the description above, we can see that KPCA is equivalent to solving the eigenvectors and eigenvalues of covariance of centered Gram matrix K.SoKPCA can be regarded as having two independent steps: kernel feature extraction and PCA on the kernel features. Dot product vector of an image is composed of kernel dot products between the image and all the training images, and then the kernel feature vector of the image is the dot product vector by subtracting its mean value. PCA is then performed on the kernel feature vectors to get the eigenvector expansion coefficient matrix A. The only difference is we need to normalize the coefficients according to (6). The projections of test data in KPCA subspace are Y =A T K test. 3 Matrix Based Kernel Feature for Image Classification Based on this new perception of KPCA, we extend the kernel feature vectors of images to kernel feature matrices to measure multiple similarities between two images. 3.1 Matrix Based Kernel Feature In KPCA, kernel dot product is used to capture the similarity between two images. But this description is not sufficient when the feature vectors of the images contain several kinds of low-level features, because it only tells the general similarity rather than individual ones in each kind of low-level visual cues. Since in image classification and retrieval, images are often represented by multiple visual cues, such as color, texture and shape, we perform kernel dot products on different visual cues respectively, and get a dot product vector between two images. This vector describes similarities in different visual cues rather than one general similarity between images. So the kernel feature vector of the image in KPCA becomes a kernel feature matrix. Assuming that there are p visual cues to represent images, the kernel feature matrix M i of image is defined as follows: k 1 (x 1 i, x1 1 ) k 2(x 2 i, x2 1 ) k p(x p i, xp 1 ) k 1 (x 1 i M i =, x1 2) k 2 (x 2 i, x2 2) k p (x p i, xp 2 ) , (10) k 1 (x 1 i, x1 N ) k 2(x 2 i, x2 N ) k p(x p i, xp N ) where k p is the dot product kernel function for the p-th visual cue, x j i is the j-th visual cue of the i-th image. We call M i the matrix based kernel feature. From the view that the dot product kernel is a similarity measure function, the matrix based kernel feature provides a multi-similarity representation, i.e., we can get p levels of similarities between the image x i and all the training images, which should be more precise than the kernel feature vector in traditional KPCA. Corresponding to the centering of vector K i in KPCA, we center each column of M i in the same way and get centered kernel feature matrix M i.

5 248 W. Yan et al. 3.2 Revised Two-Dimensional PCA Following the traditional KPCA, we have to reshape M i into a vector with N p elements first, and then perform PCA. However this reshaping lead to expensive computation due to dimension increasing by p times. For example, if there are 1000 training samples and 10 similarities between two images, then the number of dimension becomes Fortunately, the eigenvectors can be calculated efficiently using the SVD techniques [15], [16], and the process of generating the covariance matrix is actually avoided. But it is difficult to evaluate the covariance matrix accurately due to its large size and relatively small number of training samples. The eigenvectors cannot be obtained accurately, since they are determined by the covariance matrix. We revise 2DPCA algorithm proposed in [12] to deal with this problem. AsopposedtoconventionalPCA,2DPCAisbasedon2Dmatricesrather than 1D vectors. That is, the centered kernel feature matrix does not need to be previously transformed into a vector, a covariance matrix can be constructed using kernel feature matrices directly. Let {B i } N i=1 denote the training data, 2DPCA is to project B i by a transform matrix X: Z i = B i X. (11) Since the total scatter of the projected samples can be characterized by the trace of covariance matrix G of them, the following criterion is adopted to maximize the discriminating power of the projection X: 1 J(X) =trace( M 1 N (Z i µ) T (Z i µ)) = X T GX, (12) i=1 where µ is the mean matrix of all Z i s. The optimal projection X opt =[X 1, X 2,, X E ] is a matrix with each column as a unitary vector that maximize J(X), i.e., the eigenvectors of G corresponding to the first several largest eigenvalues [17]. We may take centered kernel feature matrices { } N M i as training set {B i=1 i} N i=1 and perform 2DPCA directly, or transpose { } N M i i=1 to get { } M T N i first, and i=1 perform 2DPCA. The latter scheme is adopted in this paper for its similarity to KPCA. Because each column of M i corresponds to centered kernel feature vector K i in KPCA, and the projections of the samples in KPCA subspace are the inner products between { } N K i and eigenvectors of covariance matrix C. i=1 It seems more reasonable to calculate inner products between column of M i and eigenvectors of G. Finally, the eigenvectors in X opt are normalized according to: λi (X i X i )=1, (13) where λ i is eigenvalue corresponding to eigenvector X i. For test points T =[t 1, t 2,, t L ], kernel feature matrix M test and its centered version M test can be calculated, the projections of test points are: Z = ( M test) T X. (14)

6 4 Experiments 4.1 Experimental Data Multiple Similarities Based Kernel Subspace Learning 249 We test the proposed method on the Corel image database. Our dataset contains 6000 images with 60 categories randomly selected from the Corel database. Each category with 100 manually labeled images are used as ground truth. Four kinds of visual features are used in this paper to represent the images: color histogram, color moments, wavelet based texture and orientation histogram. Color histogram is taken in HSV space with quantization of 8 4=32 bins on H and S channels. The first three moments from each of the three color channels are used for color moment. a 24-dimensional PWT based wavelet texture features and an 8-dimensional orientation histogram are contained to construct an 73-dimensional feature vector for each image. Each feature component is normalized, s.t. variance of each equals Experimental Results In our experiments, five subspace learning algorithms including the proposed method are compared. They are: PCA: This means performing PCA directly on original 73-dimensional visual features. KPCA: The Gaussian kernel is used, and we investigate the kernel parameter in [0.001, 1] and find γ =0.08 gives the best performance. The proposed method: We note it as MSPCA for simplicity. Since each image is represented by four kinds of visual cues, we perform four kernel functions to compute the kernel dot products between two images on each visual cue respectively. For each image, a matrix based kernel feature is obtained. The revised 2DPCA is applied to these features then. For simplicity, we adopt four Gaussian kernels with same parameters. By investigation of the kernel parameter in [0.001, 1], the performance of MSPCA is maximized when γ =0.5. LKPCA: As mentioned above in 3.2, after matrix based kernel features are calculated, we can reshape them into vectors first, and then perform PCA on them. The kernel functions used here are the same as in MSPCA. LPP: The code of LPP is downloaded from xiaofei/. The number of nearest neighbors N issetto10asin[4]. For the output of the above algorithms, the nearest neighbor classifier is used for classification. We test the algorithms on several different subsets of the database. Each subset is a mixture of k categories, where k varies between 2 and 10. For each category number k, 200 subsets are randomly selected from the database. Two groups of experiments are designed. The first one is designed to compare the performance of the five algorithms. For each subset, all the images are used as training data, and 75 images of each category are used as gallery, and the rest

7 250 W. Yan et al. Fig. 1. Classification results comparison among PCA, KPCA, MSPCA, LKPCA and LPP 25% images are used as probe. Fig. 1 shows the experimental results, where the classification accuracies are their average on 200 subsets. From Fig. 1, we can see that the propose algorithm, i.e., MSPCA has the best performance, followed by LKPCA, which uses the information of multiple similarities too. Since it is difficult to evaluate the covariance matrix accurately with relatively small training set in LKPCA, the accuracies are always lower than MSPCA. PCA outperforms KPCA when category number k varies from 2 to 8. Because the training set becomes more complicated with the increase of k, the performance of PCA is limited by its nature of linearity. That is why nonlinear KPCA has better results when k equals to 9 or 10. LPP fails in this experiment. In order to further evaluate the performance of the proposed method, we conduct statistical tests between the proposed method and the other four methods. Since it is hard for us to know the distributions of the accuracies, non-parametric Wilcoxon s signed rank test (one-sided) for two related samples is adopted. We conducted tests between the results of the algorithms for each category number k respectively. The null hypothesis H 0 is the result of the proposed method has the same distribution as the result of algorithm A, where A is PCA, KPCA, LKPCA or LPP. The p-value of each tests are shown in Table 1. Except for three tests (cells with italics), all p-values are less than 0.05, which means most of our tests show that there are significant differences between the accuracies of the proposed algorithm and four other algorithms respectively. Because the mean accuracies of MSPCA are always higher, MSPCA is considered better than other four algorithms at most of the time. Exceptions of the tests between PCA and MSPCA when k equals to 2 or 3 are probably due to the simplicity of training set, and adoption of kernel method such as MSPCA doesn t make much sense. But the average accuracies of MSPCA are still higher than those of PCA. The second group of experiments is to test the generalization capability of the proposed method. 50 images of each category are used as training data and as gallery, and the rest 50% are used as probe. As in the first group of experiments,

8 Multiple Similarities Based Kernel Subspace Learning 251 Table 1. P-values of hypothesis tests between the proposed method and four other algorithms respectively based on the first type of experiments. <2.2e-16 means less than 2.2e-16. k PCA KPCA LKPCA LPP e e e <2.2e e <2.2e e e e-7 <2.2e e e <2.2e e e <2.2e e e <2.2e-16 9 <2.2e e <2.2e e e <2.2e-16 Fig. 2. Generalization capability comparison among PCA, KPCA, MSPCA, LKPCA and LPP five algorithms are performed on 200 subsets and their average classification accuracies are calculated. The comparison of the result is shown in Fig. 2. These experimental results are similar to the results of the first group. MSPCA is the best, followed by LKPCA. This shows the good generalization capability of MSPCA. When the number of categories is small (no more than 8), linear PCA outperforms KPCA. For more complicated data set, KPCA is preferred to PCA. LPP also fails. We conducted Wilcoxon s signed rank tests between the results of the algorithms too, for each k respectively. The p-value of each tests are shown in Table 2. All p-values but one (the cell with italics) are small enough to show that the performance of MSPCA is better than the other four methods when using out-of-sample data, which show its good generalization capability again. When number of categories k equals 2, test between PCA and MSPCA fails to reject the null hypothesis. It is probably due to the simplicity of the image set. But the mean accuracy of MSPCA is still higher than PCA.

9 252 W. Yan et al. Table 2. P-values of hypothesis tests between the proposed method and four other algorithms respectively based on the second type of experiments k PCA KPCA LKPCA LPP <2.2e <2.2e e-5 <2.2e e-5 <2.2e e-7 <2.2e e-7 <2.2e e-12 <2.2e e-10 <2.2e-16 6 <2.2e-16 <2.2e e-8 <2.2e-16 7 <2.2e-16 <2.2e e-9 <2.2e-16 8 <2.2e-16 <2.2e e-12 <2.2e-16 9 <2.2e-16 <2.2e e-9 <2.2e <2.2e-16 <2.2e e-10 <2.2e-16 5 Conclusions In this paper, we conceive a new perception of KPCA, i.e., it can be regarded as having two separated steps: kernel features extraction and PCA based feature analysis. Dot product vector of an image is composed of kernel dot products between the image and all the training images, and then the kernel feature vector of the image is the dot product vector by subtracting its mean value. Based on this perception, we propose a new scheme of the matrix based kernel features for image clustering. With four kinds of visual cues, i.e., color histogram, color moment, wavelet based texture, and orientation histogram, we perform a dot product kernel to compute the similarity between two images respectively, and then obtain the matrix based kernel feature of an image with multi-similarities. In order to efficiently deal with the problem of the evaluation of eigenvectors, a matrix based KPCA algorithm is developed to learn the subspace of matrix features for classification. Extensive experiments on the Corel database are conducted to show the advantage of the proposed method. Acknowledgement This work is supported by the Natural Sciences Foundation of China Under Grant No , and References 1. Chen, Y., Wang, J., Krovetz, R.: Content-based image retrieval by clustering. In: Proc. of ACM SIGMM Int. Workshop on Multimedia Information Retrieval. (2003) 2. Grodon, S., Greenspan, H., Goldberger, J.: Applying the information bottleneck principal to unsupervised clustering of discrete and continuous image representations. In: Proc. of Int. Conf. Computer Vision. (2003) 3. Barreno, M.: Spectral methods for image clustering. Tech-Report CS 218B, U.C. Berkeley (2004)

10 Multiple Similarities Based Kernel Subspace Learning Zheng, X., Cai, D., He, X., Ma, W., Lin, X.: Locality preserving clustering for image database. In: Proc. of. ACM Multimedia. (2004) 5. He, X., Niyogi, P.: Locality preserving projections. In: Advances in Neural Information Processing System. Volume 16., Cambridge, MA, MIT Press (2004) 6. Schölkopf, B., Smola, A., Müller, K.: Nonlinear component analysis as a kernel eigenvalue problem. Neural Computation 10 (1998) Yang, M., Ahuja, N., Kriegman, D.: Face recognition using kernel eigenfaces. In: Proc. of. Int. Conf. Image Processing. (2000) 8. Mika, S., Schölkopf, B., Smola, A., Müller, K., Scholz, M., Rätsch, G.: Kernel PCA and de-noising in feature spaces. In: Advances in Neural Information Processing System. Volume 11., Cambridge, MA, MIT Press (1999) 9. Liu, Q., Jin, H., Tang, X., Lu, H., Ma, S.: A new perception of kernel features. Tech-Report, National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences (2005) 10. Liu, Q., Liu, H., Ma, S.: Improving kernel fisher discriminant analysis for face recognition. IEEE Trans. on Circuits and Systems for Video Technology, Special Issue on Image and Video Based Biometrics 14 (2004) Liu, Q., Huang, R., Liu, H., Ma, S.: Face recognition using kernel-based fisher discriminant analysis. In: Proc. of. Int. Conf. Automatic Face and Gesture Recognition. (2002) 12. Yang, J., Zhang, D., Frangi, A., Yang, J.: Two-Dimensional PCA: A new approach to appearance-based face representation and recognition. IEEE Trans. on Pattern Analysis and Machine Intelligence 25 (2004) Tong, S., Chang, E.: Support vector machine active learning for image retrieval. In: Proc. of. ACM Multimedia. (2001) 14. Zhang, L., Lin, F., Zhang, B.: Support vector machine for image retrieval. In: Proc. of. Int. Conf. Image Processing. (2001) 15. Sirovich, L., Kirby, M.: Low-dimensional procedure for characterization of human faces. J. Optical Soc. Am. 4 (1987) Kirby, M., Sirovich, L.: Application of the KL procedure for the characterization of human faces. IEEE Trans. on Pattern Analysis and Machine Intelligence 12 (1990) Yang, J., Yang, J.: From image vector to matrix: A straightforward image projection technique IMPCA vs. PCA. IEEE Trans. on Pattern Analysis and Machine Intelligence 35 (2002)

Connection of Local Linear Embedding, ISOMAP, and Kernel Principal Component Analysis

Connection of Local Linear Embedding, ISOMAP, and Kernel Principal Component Analysis Connection of Local Linear Embedding, ISOMAP, and Kernel Principal Component Analysis Alvina Goh Vision Reading Group 13 October 2005 Connection of Local Linear Embedding, ISOMAP, and Kernel Principal

More information

Iterative Laplacian Score for Feature Selection

Iterative Laplacian Score for Feature Selection Iterative Laplacian Score for Feature Selection Linling Zhu, Linsong Miao, and Daoqiang Zhang College of Computer Science and echnology, Nanjing University of Aeronautics and Astronautics, Nanjing 2006,

More information

SPECTRAL CLUSTERING AND KERNEL PRINCIPAL COMPONENT ANALYSIS ARE PURSUING GOOD PROJECTIONS

SPECTRAL CLUSTERING AND KERNEL PRINCIPAL COMPONENT ANALYSIS ARE PURSUING GOOD PROJECTIONS SPECTRAL CLUSTERING AND KERNEL PRINCIPAL COMPONENT ANALYSIS ARE PURSUING GOOD PROJECTIONS VIKAS CHANDRAKANT RAYKAR DECEMBER 5, 24 Abstract. We interpret spectral clustering algorithms in the light of unsupervised

More information

Local Learning Projections

Local Learning Projections Mingrui Wu mingrui.wu@tuebingen.mpg.de Max Planck Institute for Biological Cybernetics, Tübingen, Germany Kai Yu kyu@sv.nec-labs.com NEC Labs America, Cupertino CA, USA Shipeng Yu shipeng.yu@siemens.com

More information

Discriminative Direction for Kernel Classifiers

Discriminative Direction for Kernel Classifiers Discriminative Direction for Kernel Classifiers Polina Golland Artificial Intelligence Lab Massachusetts Institute of Technology Cambridge, MA 02139 polina@ai.mit.edu Abstract In many scientific and engineering

More information

Discriminant Uncorrelated Neighborhood Preserving Projections

Discriminant Uncorrelated Neighborhood Preserving Projections Journal of Information & Computational Science 8: 14 (2011) 3019 3026 Available at http://www.joics.com Discriminant Uncorrelated Neighborhood Preserving Projections Guoqiang WANG a,, Weijuan ZHANG a,

More information

Dimensionality Reduction Using PCA/LDA. Hongyu Li School of Software Engineering TongJi University Fall, 2014

Dimensionality Reduction Using PCA/LDA. Hongyu Li School of Software Engineering TongJi University Fall, 2014 Dimensionality Reduction Using PCA/LDA Hongyu Li School of Software Engineering TongJi University Fall, 2014 Dimensionality Reduction One approach to deal with high dimensional data is by reducing their

More information

Adaptive Kernel Principal Component Analysis With Unsupervised Learning of Kernels

Adaptive Kernel Principal Component Analysis With Unsupervised Learning of Kernels Adaptive Kernel Principal Component Analysis With Unsupervised Learning of Kernels Daoqiang Zhang Zhi-Hua Zhou National Laboratory for Novel Software Technology Nanjing University, Nanjing 2193, China

More information

Face Recognition Using Laplacianfaces He et al. (IEEE Trans PAMI, 2005) presented by Hassan A. Kingravi

Face Recognition Using Laplacianfaces He et al. (IEEE Trans PAMI, 2005) presented by Hassan A. Kingravi Face Recognition Using Laplacianfaces He et al. (IEEE Trans PAMI, 2005) presented by Hassan A. Kingravi Overview Introduction Linear Methods for Dimensionality Reduction Nonlinear Methods and Manifold

More information

Uncorrelated Multilinear Principal Component Analysis through Successive Variance Maximization

Uncorrelated Multilinear Principal Component Analysis through Successive Variance Maximization Uncorrelated Multilinear Principal Component Analysis through Successive Variance Maximization Haiping Lu 1 K. N. Plataniotis 1 A. N. Venetsanopoulos 1,2 1 Department of Electrical & Computer Engineering,

More information

Dominant Feature Vectors Based Audio Similarity Measure

Dominant Feature Vectors Based Audio Similarity Measure Dominant Feature Vectors Based Audio Similarity Measure Jing Gu 1, Lie Lu 2, Rui Cai 3, Hong-Jiang Zhang 2, and Jian Yang 1 1 Dept. of Electronic Engineering, Tsinghua Univ., Beijing, 100084, China 2 Microsoft

More information

Linear and Non-Linear Dimensionality Reduction

Linear and Non-Linear Dimensionality Reduction Linear and Non-Linear Dimensionality Reduction Alexander Schulz aschulz(at)techfak.uni-bielefeld.de University of Pisa, Pisa 4.5.215 and 7.5.215 Overview Dimensionality Reduction Motivation Linear Projections

More information

Nonlinear Projection Trick in kernel methods: An alternative to the Kernel Trick

Nonlinear Projection Trick in kernel methods: An alternative to the Kernel Trick Nonlinear Projection Trick in kernel methods: An alternative to the Kernel Trick Nojun Kak, Member, IEEE, Abstract In kernel methods such as kernel PCA and support vector machines, the so called kernel

More information

Efficient Kernel Discriminant Analysis via QR Decomposition

Efficient Kernel Discriminant Analysis via QR Decomposition Efficient Kernel Discriminant Analysis via QR Decomposition Tao Xiong Department of ECE University of Minnesota txiong@ece.umn.edu Jieping Ye Department of CSE University of Minnesota jieping@cs.umn.edu

More information

Kernel methods for comparing distributions, measuring dependence

Kernel methods for comparing distributions, measuring dependence Kernel methods for comparing distributions, measuring dependence Le Song Machine Learning II: Advanced Topics CSE 8803ML, Spring 2012 Principal component analysis Given a set of M centered observations

More information

Myoelectrical signal classification based on S transform and two-directional 2DPCA

Myoelectrical signal classification based on S transform and two-directional 2DPCA Myoelectrical signal classification based on S transform and two-directional 2DPCA Hong-Bo Xie1 * and Hui Liu2 1 ARC Centre of Excellence for Mathematical and Statistical Frontiers Queensland University

More information

Aruna Bhat Research Scholar, Department of Electrical Engineering, IIT Delhi, India

Aruna Bhat Research Scholar, Department of Electrical Engineering, IIT Delhi, India International Journal of Scientific Research in Computer Science, Engineering and Information Technology 2017 IJSRCSEIT Volume 2 Issue 6 ISSN : 2456-3307 Robust Face Recognition System using Non Additive

More information

DYNAMIC TEXTURE RECOGNITION USING ENHANCED LBP FEATURES

DYNAMIC TEXTURE RECOGNITION USING ENHANCED LBP FEATURES DYNAMIC TEXTURE RECOGNITION USING ENHANCED FEATURES Jianfeng Ren BeingThere Centre Institute of Media Innovation Nanyang Technological University 50 Nanyang Drive, Singapore 637553. Xudong Jiang, Junsong

More information

A Unified Bayesian Framework for Face Recognition

A Unified Bayesian Framework for Face Recognition Appears in the IEEE Signal Processing Society International Conference on Image Processing, ICIP, October 4-7,, Chicago, Illinois, USA A Unified Bayesian Framework for Face Recognition Chengjun Liu and

More information

Online Appearance Model Learning for Video-Based Face Recognition

Online Appearance Model Learning for Video-Based Face Recognition Online Appearance Model Learning for Video-Based Face Recognition Liang Liu 1, Yunhong Wang 2,TieniuTan 1 1 National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences,

More information

Statistical Machine Learning

Statistical Machine Learning Statistical Machine Learning Christoph Lampert Spring Semester 2015/2016 // Lecture 12 1 / 36 Unsupervised Learning Dimensionality Reduction 2 / 36 Dimensionality Reduction Given: data X = {x 1,..., x

More information

Nonlinear Dimensionality Reduction

Nonlinear Dimensionality Reduction Outline Hong Chang Institute of Computing Technology, Chinese Academy of Sciences Machine Learning Methods (Fall 2012) Outline Outline I 1 Kernel PCA 2 Isomap 3 Locally Linear Embedding 4 Laplacian Eigenmap

More information

Statistical Pattern Recognition

Statistical Pattern Recognition Statistical Pattern Recognition Feature Extraction Hamid R. Rabiee Jafar Muhammadi, Alireza Ghasemi, Payam Siyari Spring 2014 http://ce.sharif.edu/courses/92-93/2/ce725-2/ Agenda Dimensionality Reduction

More information

Principal Component Analysis

Principal Component Analysis CSci 5525: Machine Learning Dec 3, 2008 The Main Idea Given a dataset X = {x 1,..., x N } The Main Idea Given a dataset X = {x 1,..., x N } Find a low-dimensional linear projection The Main Idea Given

More information

Comparative Assessment of Independent Component. Component Analysis (ICA) for Face Recognition.

Comparative Assessment of Independent Component. Component Analysis (ICA) for Face Recognition. Appears in the Second International Conference on Audio- and Video-based Biometric Person Authentication, AVBPA 99, ashington D. C. USA, March 22-2, 1999. Comparative Assessment of Independent Component

More information

PCA & ICA. CE-717: Machine Learning Sharif University of Technology Spring Soleymani

PCA & ICA. CE-717: Machine Learning Sharif University of Technology Spring Soleymani PCA & ICA CE-717: Machine Learning Sharif University of Technology Spring 2015 Soleymani Dimensionality Reduction: Feature Selection vs. Feature Extraction Feature selection Select a subset of a given

More information

Lecture: Face Recognition

Lecture: Face Recognition Lecture: Face Recognition Juan Carlos Niebles and Ranjay Krishna Stanford Vision and Learning Lab Lecture 12-1 What we will learn today Introduction to face recognition The Eigenfaces Algorithm Linear

More information

below, kernel PCA Eigenvectors, and linear combinations thereof. For the cases where the pre-image does exist, we can provide a means of constructing

below, kernel PCA Eigenvectors, and linear combinations thereof. For the cases where the pre-image does exist, we can provide a means of constructing Kernel PCA Pattern Reconstruction via Approximate Pre-Images Bernhard Scholkopf, Sebastian Mika, Alex Smola, Gunnar Ratsch, & Klaus-Robert Muller GMD FIRST, Rudower Chaussee 5, 12489 Berlin, Germany fbs,

More information

Lecture 24: Principal Component Analysis. Aykut Erdem May 2016 Hacettepe University

Lecture 24: Principal Component Analysis. Aykut Erdem May 2016 Hacettepe University Lecture 4: Principal Component Analysis Aykut Erdem May 016 Hacettepe University This week Motivation PCA algorithms Applications PCA shortcomings Autoencoders Kernel PCA PCA Applications Data Visualization

More information

A METHOD OF FINDING IMAGE SIMILAR PATCHES BASED ON GRADIENT-COVARIANCE SIMILARITY

A METHOD OF FINDING IMAGE SIMILAR PATCHES BASED ON GRADIENT-COVARIANCE SIMILARITY IJAMML 3:1 (015) 69-78 September 015 ISSN: 394-58 Available at http://scientificadvances.co.in DOI: http://dx.doi.org/10.1864/ijamml_710011547 A METHOD OF FINDING IMAGE SIMILAR PATCHES BASED ON GRADIENT-COVARIANCE

More information

A Novel PCA-Based Bayes Classifier and Face Analysis

A Novel PCA-Based Bayes Classifier and Face Analysis A Novel PCA-Based Bayes Classifier and Face Analysis Zhong Jin 1,2, Franck Davoine 3,ZhenLou 2, and Jingyu Yang 2 1 Centre de Visió per Computador, Universitat Autònoma de Barcelona, Barcelona, Spain zhong.in@cvc.uab.es

More information

Example: Face Detection

Example: Face Detection Announcements HW1 returned New attendance policy Face Recognition: Dimensionality Reduction On time: 1 point Five minutes or more late: 0.5 points Absent: 0 points Biometrics CSE 190 Lecture 14 CSE190,

More information

CS4495/6495 Introduction to Computer Vision. 8C-L3 Support Vector Machines

CS4495/6495 Introduction to Computer Vision. 8C-L3 Support Vector Machines CS4495/6495 Introduction to Computer Vision 8C-L3 Support Vector Machines Discriminative classifiers Discriminative classifiers find a division (surface) in feature space that separates the classes Several

More information

The Laplacian PDF Distance: A Cost Function for Clustering in a Kernel Feature Space

The Laplacian PDF Distance: A Cost Function for Clustering in a Kernel Feature Space The Laplacian PDF Distance: A Cost Function for Clustering in a Kernel Feature Space Robert Jenssen, Deniz Erdogmus 2, Jose Principe 2, Torbjørn Eltoft Department of Physics, University of Tromsø, Norway

More information

Cluster Kernels for Semi-Supervised Learning

Cluster Kernels for Semi-Supervised Learning Cluster Kernels for Semi-Supervised Learning Olivier Chapelle, Jason Weston, Bernhard Scholkopf Max Planck Institute for Biological Cybernetics, 72076 Tiibingen, Germany {first. last} @tuebingen.mpg.de

More information

Face Recognition. Face Recognition. Subspace-Based Face Recognition Algorithms. Application of Face Recognition

Face Recognition. Face Recognition. Subspace-Based Face Recognition Algorithms. Application of Face Recognition ace Recognition Identify person based on the appearance of face CSED441:Introduction to Computer Vision (2017) Lecture10: Subspace Methods and ace Recognition Bohyung Han CSE, POSTECH bhhan@postech.ac.kr

More information

Semi-Supervised Learning through Principal Directions Estimation

Semi-Supervised Learning through Principal Directions Estimation Semi-Supervised Learning through Principal Directions Estimation Olivier Chapelle, Bernhard Schölkopf, Jason Weston Max Planck Institute for Biological Cybernetics, 72076 Tübingen, Germany {first.last}@tuebingen.mpg.de

More information

Unsupervised dimensionality reduction

Unsupervised dimensionality reduction Unsupervised dimensionality reduction Guillaume Obozinski Ecole des Ponts - ParisTech SOCN course 2014 Guillaume Obozinski Unsupervised dimensionality reduction 1/30 Outline 1 PCA 2 Kernel PCA 3 Multidimensional

More information

Principal Component Analysis

Principal Component Analysis B: Chapter 1 HTF: Chapter 1.5 Principal Component Analysis Barnabás Póczos University of Alberta Nov, 009 Contents Motivation PCA algorithms Applications Face recognition Facial expression recognition

More information

PCA, Kernel PCA, ICA

PCA, Kernel PCA, ICA PCA, Kernel PCA, ICA Learning Representations. Dimensionality Reduction. Maria-Florina Balcan 04/08/2015 Big & High-Dimensional Data High-Dimensions = Lot of Features Document classification Features per

More information

A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier

A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier A Modified Incremental Principal Component Analysis for On-Line Learning of Feature Space and Classifier Seiichi Ozawa 1, Shaoning Pang 2, and Nikola Kasabov 2 1 Graduate School of Science and Technology,

More information

Nonnegative Matrix Factorization Clustering on Multiple Manifolds

Nonnegative Matrix Factorization Clustering on Multiple Manifolds Proceedings of the Twenty-Fourth AAAI Conference on Artificial Intelligence (AAAI-10) Nonnegative Matrix Factorization Clustering on Multiple Manifolds Bin Shen, Luo Si Department of Computer Science,

More information

December 20, MAA704, Multivariate analysis. Christopher Engström. Multivariate. analysis. Principal component analysis

December 20, MAA704, Multivariate analysis. Christopher Engström. Multivariate. analysis. Principal component analysis .. December 20, 2013 Todays lecture. (PCA) (PLS-R) (LDA) . (PCA) is a method often used to reduce the dimension of a large dataset to one of a more manageble size. The new dataset can then be used to make

More information

Introduction to Machine Learning. PCA and Spectral Clustering. Introduction to Machine Learning, Slides: Eran Halperin

Introduction to Machine Learning. PCA and Spectral Clustering. Introduction to Machine Learning, Slides: Eran Halperin 1 Introduction to Machine Learning PCA and Spectral Clustering Introduction to Machine Learning, 2013-14 Slides: Eran Halperin Singular Value Decomposition (SVD) The singular value decomposition (SVD)

More information

Computer Vision Group Prof. Daniel Cremers. 2. Regression (cont.)

Computer Vision Group Prof. Daniel Cremers. 2. Regression (cont.) Prof. Daniel Cremers 2. Regression (cont.) Regression with MLE (Rep.) Assume that y is affected by Gaussian noise : t = f(x, w)+ where Thus, we have p(t x, w, )=N (t; f(x, w), 2 ) 2 Maximum A-Posteriori

More information

Kernel Uncorrelated and Orthogonal Discriminant Analysis: A Unified Approach

Kernel Uncorrelated and Orthogonal Discriminant Analysis: A Unified Approach Kernel Uncorrelated and Orthogonal Discriminant Analysis: A Unified Approach Tao Xiong Department of ECE University of Minnesota, Minneapolis txiong@ece.umn.edu Vladimir Cherkassky Department of ECE University

More information

Stefanos Zafeiriou, Anastasios Tefas, and Ioannis Pitas

Stefanos Zafeiriou, Anastasios Tefas, and Ioannis Pitas GENDER DETERMINATION USING A SUPPORT VECTOR MACHINE VARIANT Stefanos Zafeiriou, Anastasios Tefas, and Ioannis Pitas Artificial Intelligence and Information Analysis Lab/Department of Informatics, Aristotle

More information

Chemometrics: Classification of spectra

Chemometrics: Classification of spectra Chemometrics: Classification of spectra Vladimir Bochko Jarmo Alander University of Vaasa November 1, 2010 Vladimir Bochko Chemometrics: Classification 1/36 Contents Terminology Introduction Big picture

More information

Table of Contents. Multivariate methods. Introduction II. Introduction I

Table of Contents. Multivariate methods. Introduction II. Introduction I Table of Contents Introduction Antti Penttilä Department of Physics University of Helsinki Exactum summer school, 04 Construction of multinormal distribution Test of multinormality with 3 Interpretation

More information

Principal Component Analysis (PCA)

Principal Component Analysis (PCA) Principal Component Analysis (PCA) Additional reading can be found from non-assessed exercises (week 8) in this course unit teaching page. Textbooks: Sect. 6.3 in [1] and Ch. 12 in [2] Outline Introduction

More information

Bearing fault diagnosis based on EMD-KPCA and ELM

Bearing fault diagnosis based on EMD-KPCA and ELM Bearing fault diagnosis based on EMD-KPCA and ELM Zihan Chen, Hang Yuan 2 School of Reliability and Systems Engineering, Beihang University, Beijing 9, China Science and Technology on Reliability & Environmental

More information

Course 495: Advanced Statistical Machine Learning/Pattern Recognition

Course 495: Advanced Statistical Machine Learning/Pattern Recognition Course 495: Advanced Statistical Machine Learning/Pattern Recognition Deterministic Component Analysis Goal (Lecture): To present standard and modern Component Analysis (CA) techniques such as Principal

More information

Machine Learning. B. Unsupervised Learning B.2 Dimensionality Reduction. Lars Schmidt-Thieme, Nicolas Schilling

Machine Learning. B. Unsupervised Learning B.2 Dimensionality Reduction. Lars Schmidt-Thieme, Nicolas Schilling Machine Learning B. Unsupervised Learning B.2 Dimensionality Reduction Lars Schmidt-Thieme, Nicolas Schilling Information Systems and Machine Learning Lab (ISMLL) Institute for Computer Science University

More information

An Efficient Pseudoinverse Linear Discriminant Analysis method for Face Recognition

An Efficient Pseudoinverse Linear Discriminant Analysis method for Face Recognition An Efficient Pseudoinverse Linear Discriminant Analysis method for Face Recognition Jun Liu, Songcan Chen, Daoqiang Zhang, and Xiaoyang Tan Department of Computer Science & Engineering, Nanjing University

More information

Group Sparse Non-negative Matrix Factorization for Multi-Manifold Learning

Group Sparse Non-negative Matrix Factorization for Multi-Manifold Learning LIU, LU, GU: GROUP SPARSE NMF FOR MULTI-MANIFOLD LEARNING 1 Group Sparse Non-negative Matrix Factorization for Multi-Manifold Learning Xiangyang Liu 1,2 liuxy@sjtu.edu.cn Hongtao Lu 1 htlu@sjtu.edu.cn

More information

Local Fisher Discriminant Analysis for Supervised Dimensionality Reduction

Local Fisher Discriminant Analysis for Supervised Dimensionality Reduction Local Fisher Discriminant Analysis for Supervised Dimensionality Reduction Masashi Sugiyama sugi@cs.titech.ac.jp Department of Computer Science, Tokyo Institute of Technology, ---W8-7, O-okayama, Meguro-ku,

More information

Statistical and Computational Analysis of Locality Preserving Projection

Statistical and Computational Analysis of Locality Preserving Projection Statistical and Computational Analysis of Locality Preserving Projection Xiaofei He xiaofei@cs.uchicago.edu Department of Computer Science, University of Chicago, 00 East 58th Street, Chicago, IL 60637

More information

A Unified Framework for Subspace Face Recognition

A Unified Framework for Subspace Face Recognition 1222 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 26, NO. 9, SEPTEMBER 2004 A Unified Framework for Subspace Face Recognition iaogang Wang, Student Member, IEEE, and iaoou Tang,

More information

Non-linear Dimensionality Reduction

Non-linear Dimensionality Reduction Non-linear Dimensionality Reduction CE-725: Statistical Pattern Recognition Sharif University of Technology Spring 2013 Soleymani Outline Introduction Laplacian Eigenmaps Locally Linear Embedding (LLE)

More information

Learning Eigenfunctions: Links with Spectral Clustering and Kernel PCA

Learning Eigenfunctions: Links with Spectral Clustering and Kernel PCA Learning Eigenfunctions: Links with Spectral Clustering and Kernel PCA Yoshua Bengio Pascal Vincent Jean-François Paiement University of Montreal April 2, Snowbird Learning 2003 Learning Modal Structures

More information

CS4495/6495 Introduction to Computer Vision. 8B-L2 Principle Component Analysis (and its use in Computer Vision)

CS4495/6495 Introduction to Computer Vision. 8B-L2 Principle Component Analysis (and its use in Computer Vision) CS4495/6495 Introduction to Computer Vision 8B-L2 Principle Component Analysis (and its use in Computer Vision) Wavelength 2 Wavelength 2 Principal Components Principal components are all about the directions

More information

L11: Pattern recognition principles

L11: Pattern recognition principles L11: Pattern recognition principles Bayesian decision theory Statistical classifiers Dimensionality reduction Clustering This lecture is partly based on [Huang, Acero and Hon, 2001, ch. 4] Introduction

More information

Recognition Using Class Specific Linear Projection. Magali Segal Stolrasky Nadav Ben Jakov April, 2015

Recognition Using Class Specific Linear Projection. Magali Segal Stolrasky Nadav Ben Jakov April, 2015 Recognition Using Class Specific Linear Projection Magali Segal Stolrasky Nadav Ben Jakov April, 2015 Articles Eigenfaces vs. Fisherfaces Recognition Using Class Specific Linear Projection, Peter N. Belhumeur,

More information

Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method

Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method Antti Honkela 1, Stefan Harmeling 2, Leo Lundqvist 1, and Harri Valpola 1 1 Helsinki University of Technology,

More information

Clustering VS Classification

Clustering VS Classification MCQ Clustering VS Classification 1. What is the relation between the distance between clusters and the corresponding class discriminability? a. proportional b. inversely-proportional c. no-relation Ans:

More information

ADAPTIVE QUASICONFORMAL KERNEL FISHER DISCRIMINANT ANALYSIS VIA WEIGHTED MAXIMUM MARGIN CRITERION. Received November 2011; revised March 2012

ADAPTIVE QUASICONFORMAL KERNEL FISHER DISCRIMINANT ANALYSIS VIA WEIGHTED MAXIMUM MARGIN CRITERION. Received November 2011; revised March 2012 International Journal of Innovative Computing, Information and Control ICIC International c 2013 ISSN 1349-4198 Volume 9, Number 1, January 2013 pp 437 450 ADAPTIVE QUASICONFORMAL KERNEL FISHER DISCRIMINANT

More information

Motivating the Covariance Matrix

Motivating the Covariance Matrix Motivating the Covariance Matrix Raúl Rojas Computer Science Department Freie Universität Berlin January 2009 Abstract This note reviews some interesting properties of the covariance matrix and its role

More information

CITS 4402 Computer Vision

CITS 4402 Computer Vision CITS 4402 Computer Vision A/Prof Ajmal Mian Adj/A/Prof Mehdi Ravanbakhsh Lecture 06 Object Recognition Objectives To understand the concept of image based object recognition To learn how to match images

More information

Principal Component Analysis (PCA)

Principal Component Analysis (PCA) Principal Component Analysis (PCA) Salvador Dalí, Galatea of the Spheres CSC411/2515: Machine Learning and Data Mining, Winter 2018 Michael Guerzhoy and Lisa Zhang Some slides from Derek Hoiem and Alysha

More information

ECE 521. Lecture 11 (not on midterm material) 13 February K-means clustering, Dimensionality reduction

ECE 521. Lecture 11 (not on midterm material) 13 February K-means clustering, Dimensionality reduction ECE 521 Lecture 11 (not on midterm material) 13 February 2017 K-means clustering, Dimensionality reduction With thanks to Ruslan Salakhutdinov for an earlier version of the slides Overview K-means clustering

More information

Machine Learning (BSMC-GA 4439) Wenke Liu

Machine Learning (BSMC-GA 4439) Wenke Liu Machine Learning (BSMC-GA 4439) Wenke Liu 02-01-2018 Biomedical data are usually high-dimensional Number of samples (n) is relatively small whereas number of features (p) can be large Sometimes p>>n Problems

More information

A SEMI-SUPERVISED METRIC LEARNING FOR CONTENT-BASED IMAGE RETRIEVAL. {dimane,

A SEMI-SUPERVISED METRIC LEARNING FOR CONTENT-BASED IMAGE RETRIEVAL. {dimane, A SEMI-SUPERVISED METRIC LEARNING FOR CONTENT-BASED IMAGE RETRIEVAL I. Daoudi,, K. Idrissi, S. Ouatik 3 Université de Lyon, CNRS, INSA-Lyon, LIRIS, UMR505, F-696, France Faculté Des Sciences, UFR IT, Université

More information

Immediate Reward Reinforcement Learning for Projective Kernel Methods

Immediate Reward Reinforcement Learning for Projective Kernel Methods ESANN'27 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), 25-27 April 27, d-side publi., ISBN 2-9337-7-2. Immediate Reward Reinforcement Learning for Projective Kernel Methods

More information

Advanced Introduction to Machine Learning CMU-10715

Advanced Introduction to Machine Learning CMU-10715 Advanced Introduction to Machine Learning CMU-10715 Principal Component Analysis Barnabás Póczos Contents Motivation PCA algorithms Applications Some of these slides are taken from Karl Booksh Research

More information

STUDY ON METHODS FOR COMPUTER-AIDED TOOTH SHADE DETERMINATION

STUDY ON METHODS FOR COMPUTER-AIDED TOOTH SHADE DETERMINATION INTERNATIONAL JOURNAL OF INFORMATION AND SYSTEMS SCIENCES Volume 5, Number 3-4, Pages 351 358 c 2009 Institute for Scientific Computing and Information STUDY ON METHODS FOR COMPUTER-AIDED TOOTH SHADE DETERMINATION

More information

Kernel Methods. Barnabás Póczos

Kernel Methods. Barnabás Póczos Kernel Methods Barnabás Póczos Outline Quick Introduction Feature space Perceptron in the feature space Kernels Mercer s theorem Finite domain Arbitrary domain Kernel families Constructing new kernels

More information

Face recognition Computer Vision Spring 2018, Lecture 21

Face recognition Computer Vision Spring 2018, Lecture 21 Face recognition http://www.cs.cmu.edu/~16385/ 16-385 Computer Vision Spring 2018, Lecture 21 Course announcements Homework 6 has been posted and is due on April 27 th. - Any questions about the homework?

More information

Machine Learning - MT & 14. PCA and MDS

Machine Learning - MT & 14. PCA and MDS Machine Learning - MT 2016 13 & 14. PCA and MDS Varun Kanade University of Oxford November 21 & 23, 2016 Announcements Sheet 4 due this Friday by noon Practical 3 this week (continue next week if necessary)

More information

Unsupervised Learning Techniques Class 07, 1 March 2006 Andrea Caponnetto

Unsupervised Learning Techniques Class 07, 1 March 2006 Andrea Caponnetto Unsupervised Learning Techniques 9.520 Class 07, 1 March 2006 Andrea Caponnetto About this class Goal To introduce some methods for unsupervised learning: Gaussian Mixtures, K-Means, ISOMAP, HLLE, Laplacian

More information

Symmetric Two Dimensional Linear Discriminant Analysis (2DLDA)

Symmetric Two Dimensional Linear Discriminant Analysis (2DLDA) Symmetric Two Dimensional inear Discriminant Analysis (2DDA) Dijun uo, Chris Ding, Heng Huang University of Texas at Arlington 701 S. Nedderman Drive Arlington, TX 76013 dijun.luo@gmail.com, {chqding,

More information

Approximate Kernel PCA with Random Features

Approximate Kernel PCA with Random Features Approximate Kernel PCA with Random Features (Computational vs. Statistical Tradeoff) Bharath K. Sriperumbudur Department of Statistics, Pennsylvania State University Journées de Statistique Paris May 28,

More information

A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier

A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier A Modified Incremental Principal Component Analysis for On-line Learning of Feature Space and Classifier Seiichi Ozawa, Shaoning Pang, and Nikola Kasabov Graduate School of Science and Technology, Kobe

More information

Linear Dimensionality Reduction

Linear Dimensionality Reduction Outline Hong Chang Institute of Computing Technology, Chinese Academy of Sciences Machine Learning Methods (Fall 2012) Outline Outline I 1 Introduction 2 Principal Component Analysis 3 Factor Analysis

More information

Linear Algebra Methods for Data Mining

Linear Algebra Methods for Data Mining Linear Algebra Methods for Data Mining Saara Hyvönen, Saara.Hyvonen@cs.helsinki.fi Spring 2007 Linear Discriminant Analysis Linear Algebra Methods for Data Mining, Spring 2007, University of Helsinki Principal

More information

CSC 411 Lecture 12: Principal Component Analysis

CSC 411 Lecture 12: Principal Component Analysis CSC 411 Lecture 12: Principal Component Analysis Roger Grosse, Amir-massoud Farahmand, and Juan Carrasquilla University of Toronto UofT CSC 411: 12-PCA 1 / 23 Overview Today we ll cover the first unsupervised

More information

Lecture: Face Recognition and Feature Reduction

Lecture: Face Recognition and Feature Reduction Lecture: Face Recognition and Feature Reduction Juan Carlos Niebles and Ranjay Krishna Stanford Vision and Learning Lab Lecture 11-1 Recap - Curse of dimensionality Assume 5000 points uniformly distributed

More information

Machine Learning. Dimensionality reduction. Hamid Beigy. Sharif University of Technology. Fall 1395

Machine Learning. Dimensionality reduction. Hamid Beigy. Sharif University of Technology. Fall 1395 Machine Learning Dimensionality reduction Hamid Beigy Sharif University of Technology Fall 1395 Hamid Beigy (Sharif University of Technology) Machine Learning Fall 1395 1 / 47 Table of contents 1 Introduction

More information

Color VQ-Based Image Compression by Manifold Learning

Color VQ-Based Image Compression by Manifold Learning Color VQ-Based Image Compression by Manifold Learning Christophe Charrier 1,2 and Olivier Lézoray 1 1 Université de Caen Basse-Normandie Laboratoire GREYC, Unité Mixte de Recherche CNRS 6072 6 Bd. Maréchal

More information

COMS 4721: Machine Learning for Data Science Lecture 19, 4/6/2017

COMS 4721: Machine Learning for Data Science Lecture 19, 4/6/2017 COMS 4721: Machine Learning for Data Science Lecture 19, 4/6/2017 Prof. John Paisley Department of Electrical Engineering & Data Science Institute Columbia University PRINCIPAL COMPONENT ANALYSIS DIMENSIONALITY

More information

Kernel Methods. Charles Elkan October 17, 2007

Kernel Methods. Charles Elkan October 17, 2007 Kernel Methods Charles Elkan elkan@cs.ucsd.edu October 17, 2007 Remember the xor example of a classification problem that is not linearly separable. If we map every example into a new representation, then

More information

Two-Layered Face Detection System using Evolutionary Algorithm

Two-Layered Face Detection System using Evolutionary Algorithm Two-Layered Face Detection System using Evolutionary Algorithm Jun-Su Jang Jong-Hwan Kim Dept. of Electrical Engineering and Computer Science, Korea Advanced Institute of Science and Technology (KAIST),

More information

Eigenimages. Digital Image Processing: Bernd Girod, 2013 Stanford University -- Eigenimages 1

Eigenimages. Digital Image Processing: Bernd Girod, 2013 Stanford University -- Eigenimages 1 Eigenimages " Unitary transforms" Karhunen-Loève transform" Eigenimages for recognition" Sirovich and Kirby method" Example: eigenfaces" Eigenfaces vs. Fisherfaces" Digital Image Processing: Bernd Girod,

More information

Introduction to Machine Learning

Introduction to Machine Learning 10-701 Introduction to Machine Learning PCA Slides based on 18-661 Fall 2018 PCA Raw data can be Complex, High-dimensional To understand a phenomenon we measure various related quantities If we knew what

More information

Kazuhiro Fukui, University of Tsukuba

Kazuhiro Fukui, University of Tsukuba Subspace Methods Kazuhiro Fukui, University of Tsukuba Synonyms Multiple similarity method Related Concepts Principal component analysis (PCA) Subspace analysis Dimensionality reduction Definition Subspace

More information

Reconnaissance d objetsd et vision artificielle

Reconnaissance d objetsd et vision artificielle Reconnaissance d objetsd et vision artificielle http://www.di.ens.fr/willow/teaching/recvis09 Lecture 6 Face recognition Face detection Neural nets Attention! Troisième exercice de programmation du le

More information

Expectation Maximization

Expectation Maximization Expectation Maximization Machine Learning CSE546 Carlos Guestrin University of Washington November 13, 2014 1 E.M.: The General Case E.M. widely used beyond mixtures of Gaussians The recipe is the same

More information

ESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN ,

ESANN'2001 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), April 2001, D-Facto public., ISBN , Sparse Kernel Canonical Correlation Analysis Lili Tan and Colin Fyfe 2, Λ. Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong. 2. School of Information and Communication

More information

Data Mining. Dimensionality reduction. Hamid Beigy. Sharif University of Technology. Fall 1395

Data Mining. Dimensionality reduction. Hamid Beigy. Sharif University of Technology. Fall 1395 Data Mining Dimensionality reduction Hamid Beigy Sharif University of Technology Fall 1395 Hamid Beigy (Sharif University of Technology) Data Mining Fall 1395 1 / 42 Outline 1 Introduction 2 Feature selection

More information

Distance Metric Learning in Data Mining (Part II) Fei Wang and Jimeng Sun IBM TJ Watson Research Center

Distance Metric Learning in Data Mining (Part II) Fei Wang and Jimeng Sun IBM TJ Watson Research Center Distance Metric Learning in Data Mining (Part II) Fei Wang and Jimeng Sun IBM TJ Watson Research Center 1 Outline Part I - Applications Motivation and Introduction Patient similarity application Part II

More information

Deriving Principal Component Analysis (PCA)

Deriving Principal Component Analysis (PCA) -0 Mathematical Foundations for Machine Learning Machine Learning Department School of Computer Science Carnegie Mellon University Deriving Principal Component Analysis (PCA) Matt Gormley Lecture 11 Oct.

More information