Combining evidence in personal identity verification systems
|
|
- Silas Eaton
- 6 years ago
- Views:
Transcription
1 Pattern Recognition Letters ELSEVIER Pattern Recognition Letters 18 (1997) Combining evidence in personal identity verification systems J. Kittler *, J. Matas 1, K. Jonsson, M.U. Ramos S~nchez University of Surrey, Guildford GU2 5XH, United Kingdom Abstract A methodology for fusing multiple instances of biometric data to improve the performance of a personal identity verification system is developed. The fusion problem is formulated in the framework of the Bayesian estimation theory. The effect of different fusion strategies on the error probability is analysed theoretically. The proposed methodology is then demonstrated on the problem of personal identity verification using multiple facial images. Experimental studies on the M2VTS database confirm the predicted improvements in performance. A reduction in error rates of up to 40% is achieved. The performance gains are initially monotonic but they tend to saturate after integrating the first few observations. It is also shown that the fusion based on rank order statistic, i.e., the median, is robust to outliers Elsevier Science B.V. Keywords: Evidence combination; Face identification; Multiple observation fusion; Error sensitivity I. Introduction The ultimate goal of designing a personal identity verification system is to achieve the best possible performance. Traditionally, the decision making process would draw on a single modality of biometric information. However, recent studies show that the fusion of multiple sources of evidence is likely to lead to tangible benefits in terms of improved performance and robustness of the verification system. Several different sources of information can be brought to bear on the verification task. In the first instance one can employ multiple experts to provide opinions about the same biometric data. Such approaches have been shown theoretically to improve recognition rates both for discriminant function (Tumer and Ghosh, 1996) and Bayesian (Kittler et * Corresponding author. j.kittler@ee.surrey.ac.uk. J g.matas@ee.surrey.ac.uk. al., 1997) decision rules. Experimental evidence of performance gains that can be achieved by expert fusion has been reported in (Genoud et al., 1996) in the context of voice based speaker verification. Three verification methods are used to compare voice references extracted from the speech signal represented by Linear Prediction Coefficients. The global decision is taken by applying a Furui threshold to the individual methods and then combining the individual results according to a majority rule. The fusion of multimodal biometric data is exemplified by (Brunelli and Falavigna, 1995; Brunelli et al., 1995; Bigiin et al., 1997; Jourlin et al., 1997). For instance the person identification system developed by Brunelli (Brunelli and Falavigna, 1995; Brunelli et al., 1995) is based on acoustic and visual features. The system is organised as a set of non-homogeneous classifiers whose outputs are integrated after a normalisation step. In particular, two classifiers based on acoustic features and three based on visual ones provide data for an integration module /97/$ Elsevier Science B.V. All rights reserved. PII SO (97)
2 846 J. Kittler et al. / Pattern Recognition Letters 18 (1997) whose performance is shown to be superior to that of the acoustic and visual subsystems. Another system based on geometrical facial features and achieving similar results can be found in (Brunelli et al., 1995). Although the face recognition performance using geometrical features was slightly worse than that obtained using a template matching (Brunelli and Poggio, 1993), face representation is more compact and less computationally demanding, thus providing an example of integration where simple and fast recognition modules are combined. In this paper we investigate the benefits of fusing multiple measurements on a single biometric modality using a single expert. In particular we use a recently developed person identification system based on frontal face image verification (Matas et al., 1997). The system is applied to a number of images acquired during a dialogue with the prospective user and the final decision is made by fusing the soft decision outputs for the individual frames. The fusion problem is formulated for different combination strategies in the framework of Bayesian estimation theory. We analyse the effect of fusing multiple observations on error probability theoretically and confirm the predicted improvements experimentally. We show that the performance gains achieved by means of fusion are initially monotonic but soon reach saturation (after three or four frames in our experiments). It is also shown that fusion rules based on rank order statistics are robust to outliers. In summary, the contribution of the paper is twofold. First of all, we develop a methodology for fusing multiple instances of biometric data to ameliorate the verification system performance. We demonstrate the proposed methodology on the problem of personal identity verification using facial images and confirm the theoretical prediction of performance gains experimentally on the M2VTS database (Pigeon and Vandendrope, 1997). 2 The paper is organised as follows. In Section 2 we formulate the evidence combination problem and introduce the necessary notation. In this section we also present various information fusion strategies. Section 3 investigates the effect of a typical integra- 2 This database is in the public domain. tion scheme, fusion by averaging, on the verification errors. Section 4 describes the face verification system used in the experimental studies of the advocated fusion strategies. In Section 5 we present the results of experiments which show up to 40% improvement in performance. Finally, Section 6 summarises the main results of the paper and offers concluding remarks. 2. Combination strategies Consider a personal identity verification problem where user Z is claiming access to some services or a facility. The claimed identity can be either accepted or rejected. We do not consider the semi-automatic scenario where ask f o r assistance decision may be taken in borderline cases. In other words the user is to be assigned to one of the two possible classes {o~,,w 2} where w, means accept and oj 2 signifies an impostor. The user is represented by a measurement on a biometric modality, such as a face image. We assume that the verification system will engage the user in a dialogue over a short period of time and that during this period it will be possible to acquire several instances of the client description for decision making. Let us assume that we have R instances of the biometric measurement vector x i, i = R. In the measurement space each class w k is modelled by the probability density function p ( x i] w k) and its apriori probability of occurrence is denoted P ( w k ). We shall consider the models to be mutually exclusive. For a single instance xi the decision to accept or reject would be based on the aposteriori class probabilities P ( w j l x i ), j = 1,2, However we wish to take advantage of the additional measurement information and base the decision on all the instances available. In order to investigate the benefit of the additional information, let us examine possible fusion models that could be adopted. Note that the successive measurements x i are acquired within a short interval of time and therefore under the same measurement conditions. We can therefore consider x~s as repeated (multiple) measurements which will differ from each other only in terms of their noise components. In other words we can view each observation
3 J. Kittler et al. / Pattern Recognition Letters 18 (1997) as a corrupted version of some nominal value x. Thus the aposteriori class probabilities based on these observations can be considered as noisy estimates of the nominal value P(coflx), i.e., P(coflxi) = P(cojlx) + e(co, lxi), (1) where 6(co, Ix i) denotes the estimation error associated with the ith instance. For simplicity we shall assume that irrespective of class the estimation errors are identically distributed with a zero mean and variance Oe 2. This formulation of the verification problem immediately suggests that better estimates of the aposteriori class probabilities P(cojlx), j = 1,2 can be obtained by a straightforward application of the estimation theory. Accordingly, the estimates can be combined linearly or using rank order statistics to obtain an improved estimate /~(coilx). This then leads to the following combination strategies: Averaging 1 R /3(co, Ix) = ~- EP(co, lxi). (2) i=l Max rule fi( co, Ix) = max 5,P( coklxi). (3) Min rule fi( coflx ) = min~=, P( cokix i)" (4) Median rule fi( co, Ix ) = med~=, P( coklx i ). (5) In the following section we shall investigate the effect of such improved estimates on the system verification error. 3. Effect of fusion Since the estimation errors are assumed to be unbiased, the combined estimate fi(co, lx) will also be unbiased, i.e., E( fi( cojlx ) } = P( cojlx ) (6) and its variance will be reduced depending on the combination strategy used. For the averaging combi- nation, assuming the errors are uncorrelated, the variance reduces to ~ 2 = o.ez/r. Let us, for the moment, confine our discussion to the averaging combiner and explore how this reduced variance impacts on the error probability. To this end we have to establish what the probability is for the recognition system to make an error which will exceed the Bayes error defined by the nominal posteriors P(coflx). Suppose that P(collX)> P(co2[x). The Bayes error will be exceeded when the aposteriori class probability estimate for class co2 becomes maximum. Let us derive the probability of the event fi(co21x) - / ; ( c o, I x ) > 0 (7) occurring. Note that the left-hand side of Eq. (7) can be expressed as P(co2lx) - P(tOllX) + e(co2tx) - e(collx) > 0, (8) where e(cojlx) is the average error after fusion. For the averaging combiner the error after fusion is given as 1 n ~(cojix) = -'e i ~ 1 ~ ( cojlx i ). (9) Eq. (8) defines a constraint for the two estimation errors e(coflx), j = 1,2 as o ( co21x)- e ( coil X ) > P ( col IX ) -- P(co21x). (10) Note that the errors on the left-hand side of Eq. (10) satisfy s ( c o 2 1 x ) = - e ( c o l l x ). Let us assume that the distribution of the random variable e(coilx) is Gaussian with standard deviation 6". This in practice will approximate the true distribution of estimation errors very coarsely, as both ends of the [0,1 ] interval from which the aposteriori class probabilities can assume values, will clip the errors. Nevertheless, the analysis under even such a simplistic assumption will give an indication of the benefits of fusion of multiple observations. Since the error distribution is Gaussian, the distribution of the difference of the two dependent random variables will also be Gaussian with a twice as large standard deviation. The probability of the constraint in Eq. (10) being satisfied is given by the area
4 848 J. Kittler et al. / Pattern Recognition Letters 18 (1997) under the Gaussian tail with a cut off point at P ( w t l x ) - P(wzlX). More specifically, this probability, which we shall denote Q21(A P12(x)), is given by o:,(ae,:(x)) AP'2(x) ] AP,2(x ) > 0, (11) 2 V~- 6- ] where A P I 2 ( X ) = P ( t o l l x ) - P ( w 2 l x ), and erf(ap12(x)/2v~6") is the error function defined as erf(~p~2(x)2v~-~ ) ~[AP'2(x)a'Yn-(I/2Xv2/4d'2)d T. (12) Thus the client will be misclassified as an impostor with probability Q21(APx2(x)). By analogy, an impostor will be accepted as a client with probability Q12(A P21(x)). Let us denote the additional (over and above the Bayes error) probability by Q(x), i.e., Q(x) ={ Q21(AP12(X)) i f P ( t o, l x ) > P ( w 2 I x ), O,2(AP2,(x)) i f P ( t o l l x ) ~ e ( w 2 l x ). (13) The additional error probability Q(x) will be significant only in the regions where the two aposteriori class probabilities are comparable as a large positive argument of Qij(APji(x)) will drive the additional probability to zero. The average additional misclassification error will then be ~= fq(x)p(x)dx, (14) where p(x) is the mixture density. Recalling Eq. (11), each probability Q i j ( A e j i ( x ) ) in Eq. (13) depends heavily on the variance of the error of the aposteriori class probability estimate. With the multiplicity of observations increasing, the estimate variance will be reduced commensurately. However, the probability of the additional error goes down much more dramatically. In comparison with a single in- stance R = 1, the probability of the pointwise error, for APji(x) > 0 is reduced by a factor 1 - erf( A pji( x ) /2x/2O-e) l _ erf( A ei,( x ) ( ) ) (15) Note that these improvements are achieved only near the decision boundaries, as far from the boundaries the probability of a pattern x being misclassified is negligible. Thus these impressive improvements will be diluted by the averaging process in Eq. (14) where over large regions the local probability of additional error will be effectively zero because of the large difference between the two class aposteriori probabilities. However, whatever the final improvement is, any enhancement in the system performance, especially when one operates close to the 100% mark will be welcome and may make all the difference between an exercise of theoretical interest and a commercially viable solution. Similarly, one can derive the variance reduction gains for rank order statistics combination rules, such as max, min and median rules. The gains can be shown to depend on the number of instances R, the shape of the estimation error distribution and also on the rank order. Although the gains are, on the whole, less dramatic as compared with the averaging combiner, they are more robust to outliers. 4. Optimised robust correlation The face verification system we used to evaluate the different combination strategies is based on optimised robust correlation (Matas et al., 1997). The objective of this method is to find the global extremum in a multi-dimensional search space that corresponds to the best match between a pair of images. This search space is defined by the set of all valid geometric and photometric transformations. In our implementation of the optimised robust correlation the geometric transformations are translation, scaling and rotation. Below we present the score function used to evaluate a point in the multi-dimensional search space, the search method employed to find the global optimum of the score function and, finally, a sampling technique used to speed up the correlation.
5 J. Kittler et al. / Pattern Recognition Letters 18 (1997) I I I I I Fig. 1. Sample images from the M2VTS database illustrating changes in the appearance of a client Score function Given a transformation t, a match score s is computed as a weighted sum of two scores, an area score Sarea and a grey level score Sgrey: s(t) = a. Sarea(t ) q- (1 -- or) " Sgrey(t), where a denotes a constant in the interval [0,1]. The area score is included to encourage large overlaps and is defined as Iar n Stl- Iatl S.... (l)- iarl+larnatl, where S r and S t denote the sampling sets corresponding to the reference and test images, I r and I t, respectively. The grey level score, which measures the similarity between the intensity distributions, is defined as Sgrey(t ) = E f k ( f i ( l r ( P r ) ), I t ( f p ( t ' P r ) ) ) Pr~SrnSt f~nax IS r n Stl where fk denotes a robust kernel, f~ a linear intensity transformation, fp an affine projection function and f~nax the maximum response of the robust kernel. Refer to (Matas et al., 1997) for a more detailed discussion of the score function. of the search, the transformation between reference and test image is perturbed by adding a random vector drawn from an exponential distribution. The new transformation is accepted only if the score was increased. This optimisation method is similar to simulated annealing (Kirkpatrick et al., 1983) at zero temperature Random sampling A sampling technique commonly used in Monte Carlo integration is based on Sobol sequences (Press et al., 1992). These quasi-random sequences allow faster convergence compared to uniformly distributed random numbers since the fractional error of the approximation decreases as ln(n)d/n instead of 1 / v ~ (Press et al., 1992), where N is the number of samples and d the dimensionality of the approximated function. Combining the optimised robust correlation method with random sampling yields several benefits: the execution time can be significantly reduced since only a fraction of the pixels is needed, the quasi-random nature of the process prevents aliasing and it is possible to continuously increase the sampling density and at the same time maintain approximately uniform density throughout the image Optimisation method 5. Experimental results The search technique we employ is based on random exponential perturbations. In each iteration The experiments summarised below were all performed on images from the M2VTS multi-modal,, Fig. 2. An example of a client test: Person JR shot 1 against shot 3.
6 850 J. Kittler et al. / Pattern Recognition Letters 18 (1997) &) O.lO 0.08 " i i ' ~ T ( ' - ' - " ~... T O,OB i.i i,. ~ i i L. "~ ~"i i ~... i i ~ k- ' t..-+. " i t ~ * i : ~,,e~ t!...~ i...i..._.~...i , i! ' i i - - ' ~ = i i! ~-~-.o s o.08 I--...r-...l--.-.%~... N - ~ ~ : ~ ~ 0.02 O'OCo.o , False Acceptance?. ~ I.... i z : FaIN Acceptance ) o _ ~o.o4... i... ~ ~ ~ L [ i i i i.i. lil I,i 0"0~ , False Acceptance Fig. 3. (a) Comparison between different combination strategies (2 models per shot) and verification performance as a function of number of test frames using (b) average and (c) median for combination (1 model per lip state and shot). database (Pigeon and Vandendrope, 1997). This publicly available database contains facial images and recordings of speech from 37 persons. For each person, 5 "shots" 3 acquired over a period of several weeks are available. A single shot is made up of 3 sequences: (1) a frontal-view sequence in which the person is counting from 0 to 9, (2) a rotation sequence in which the person is moving his or her head and (3) a rotation sequence identical to the previous one except that, if glasses are present, they are removed. Some sample images from the M2VTS database are shown in Fig. 1. An example of a client test output from the experiments reported below is shown in Fig. 2. The interleaved image in Fig. 2(c) was obtained by transforming the reference image and then selecting rows interchangeably from the transformed image and the test image. The response image in Fig. 2(d) was computed by applying the robust kernel to each pixel 3 A take is called a shot in the M2VTS terminology. in the overlapping region between the transformed reference image and the test image. To illustrate the benefits of the different combination strategies an experiment was performed using images from one of the two rotation sequences of the first four shots. The test set contained both frontalview and non-frontal-view images whereas the reference set contained frontal-view images only. For each test image, the maximum score obtained over all reference images were taken. This effectively corresponds to a nearest-neighbour classification. The total score for each sequence of test images was then computed by using the four different combination rules described in Section 2. The verification performance was estimated using the l e a v e - o n e - o u t methodology in which training and testing sets are disjoint. The receiver operating characteristics (ROC) are shown in Fig. 3(a). The equal error rates (EERs) for the product, median, average and maximum rules are 5.5, 5.1, 5.5 and 4.6%, respectively. As can be seen in Fig. 3(a), the combination rules based on rank order statistics, the median and maximum rules,
7 J. Kittler et al. / Pattern Recognition Letters 18 (1997) Table l Equal error rate as a function of number of test frames using average (A) and median (M) for combination No. frames EER, A EER, M reduction in error rates of up to 40% was achieved. We showed that the performance gains are initially monotonic but they tend to saturate after integrating the first few observations. It was also shown that the fusion based on a rank order statistics, i.e., the median, is robust to outliers. perform better which is a direct consequence of their robustness to outliers (such as non-frontal views). In order to investigate the relationships between verification performance and number of test frames used an experiment was performed using images from the frontal-view sequences of the first four shots. A chromaticity-based lip tracker described in (Ramos et al., 1997) was used to select three "open-mouth" and three "shut-mouth" images from each shot. 4 The ROC curves obtained by varying the number of test frames from one to six and using average and median for combination are shown in Fig. 3(b) and (c), respectively. The corresponding EERs are listed in Table 1. Note the reduction in error rate for the first three or four frames followed by a saturation in recognition performance. The average outperforms the median which is mainly due to the absence of outliers in the dataset and the statistical superiority of the average combination rule in these situations. 6. Conclusions A methodology for fusing multiple instances of biometric data to improve the performance of a personal identity verification system has been developed. The fusion problem was formulated in the framework of the Bayesian estimation theory. We analysed the effect of different fusion strategies on the error probability theoretically. We then demonstrated the proposed methodology on the problem of personal identity verification using facial images and confirmed the predicted improvements in performance experimentally on the M2VTS database. A 4 The use of a lip tracker to determine the state of the mouth facilitates model selection with a resulting reduction in population entropy (Kittler et al., 1997). Acknowledgements This work was carried out as part of the European Commission ACTS Project M2VTS and was partially supported by ESPRIT Projects SAM (contract ECIS003) and RETINA. References Bigiin, E.S., Bigiin, J., Duc, B., Fischer, S., Expert conciliation for multi modal person authentication systems by bayesian statistics. In: Bigun, J., Chollet, G., Borgefors, G. (Eds.), Audio- and Video-based Biometric Person Authentication. Lecture Notes in Computer Science, vol Springer, Berlin, pp Brunelli, R., Falavigna, D., Person identification using multiple cues. IEEE Trans. Pattern Anal. Machine Intell. 17 (10), Brunelli, R., Falavigna, D., Poggio, T., Stringa, L., Automatic person recognition by using acoustic and geometric features. Machine Vision and Applications 8, Brunelli, R., Poggio, T., Face recognition: Features versus templates. IEEE Trans. Pattern Anal. Machine Intell. 15 (10), Genoud, D., Gravier, G., Bimbot, F., Chollet, G., Combining methods to improve the phone based speaker verification decision. In: Chollet, G., BigUn, J., Borgefors, G. (Eds.) ICSLP'96, vol. 3, pp Jourlin, P., Luettin, J., Genoud, D., Wassner, H., Acousticlabial speaker verification. In: Chollet, G., Bigiin, J., Borgefors, G. (Eds.), Lecture Notes in Computer Science, vol Springer, Berlin, pp Kirkpatrick, S., Gelatt, C.D., Vecchi, M.P., Optimization by simulated annealing. Science 220 (4598), Kittler, J., Li, Y.P., Matas, J., Ramos Sfinchez, M.U., Lip-shape dependent face verification. In: Bigun, J., Chollet, G., Borgefors, G. (Eds.), Audio- and Video-based Biometric Person Authentication. Lecture Notes in Computer Science, vol Springer, Berlin, pp Matas, J., Jonsson, K., Kittler, J., Fast face localisation and verification. In: Clark, A. (Ed.), British Machine Vision Conference, BMVA Press. Pigeon, S., Vandendrope, L., The m2vts multimodal face database (release 1.00). In: Bigun, J., Chollet, G., Borgefors,
8 852 J. Kittler et al. / Pattern Recognition Letters 18 (1997) G. (Eds.), Audio- and Video-based Biometric Person Authentication, Lecture Notes in Computer Science, vol Springer, Berlin, pp Press, W.H., Flannery, B.P., Teukolsky, S.A., Vetterling, W.T., Numerical Recipes in C. Cambridge University Press. Ramos, M.U., S~nchez, Matas, J., Kittler, J., Statistical chromaticity-based lip tracking with b-splines. In: Proc. Internat. Conf. on Acoustics, Speech and Signal Processing, Mu- nich, Germany, April 21-24, vol. 4. IEEE Computer Society P r e s s, p p S e e a l s o : /lip_tracking/index.html. Tumer, K., Ghosh, J., Analysis of decision boundaries in linearly combined neural classifiers. Pattern Recognition 29 (2),
I D I A P R E S E A R C H R E P O R T. Samy Bengio a. November submitted for publication
R E S E A R C H R E P O R T I D I A P Why Do Multi-Stream, Multi-Band and Multi-Modal Approaches Work on Biometric User Authentication Tasks? Norman Poh Hoon Thian a IDIAP RR 03-59 November 2003 Samy Bengio
More informationLinear Combiners for Fusion of Pattern Classifiers
International Summer School on eural ets duardo R. Caianiello" SMBL MTHODS FOR LARIG MACHIS Vietri sul Mare(Salerno) ITALY 22-28 September 2002 Linear Combiners for Fusion of Pattern Classifiers Lecturer
More informationDynamic Linear Combination of Two-Class Classifiers
Dynamic Linear Combination of Two-Class Classifiers Carlo Lobrano 1, Roberto Tronci 1,2, Giorgio Giacinto 1, and Fabio Roli 1 1 DIEE Dept. of Electrical and Electronic Engineering, University of Cagliari,
More informationInformation Fusion for Local Gabor Features Based Frontal Face Verification
Information Fusion for Local Gabor Features Based Frontal Face Verification Enrique Argones Rúa 1, Josef Kittler 2, Jose Luis Alba Castro 1, and Daniel González Jiménez 1 1 Signal Theory Group, Signal
More informationNearest Neighbourhood Classifiers in Biometric Fusion
Nearest Neighbourhood Classifiers in Biometric Fusion Nearest Neighbourhood Classifiers in Biometric Fusion A. Teoh, S. A. Samad, and A. Hussain Electrical, Electronic and System Engineering Department
More informationLecture 16: Small Sample Size Problems (Covariance Estimation) Many thanks to Carlos Thomaz who authored the original version of these slides
Lecture 16: Small Sample Size Problems (Covariance Estimation) Many thanks to Carlos Thomaz who authored the original version of these slides Intelligent Data Analysis and Probabilistic Inference Lecture
More informationComputer Vision Group Prof. Daniel Cremers. 10a. Markov Chain Monte Carlo
Group Prof. Daniel Cremers 10a. Markov Chain Monte Carlo Markov Chain Monte Carlo In high-dimensional spaces, rejection sampling and importance sampling are very inefficient An alternative is Markov Chain
More informationBIOMETRIC verification systems are used to verify the
86 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 14, NO. 1, JANUARY 2004 Likelihood-Ratio-Based Biometric Verification Asker M. Bazen and Raymond N. J. Veldhuis Abstract This paper
More informationRapid Object Recognition from Discriminative Regions of Interest
Rapid Object Recognition from Discriminative Regions of Interest Gerald Fritz, Christin Seifert, Lucas Paletta JOANNEUM RESEARCH Institute of Digital Image Processing Wastiangasse 6, A-81 Graz, Austria
More informationBiometrics: Introduction and Examples. Raymond Veldhuis
Biometrics: Introduction and Examples Raymond Veldhuis 1 Overview Biometric recognition Face recognition Challenges Transparent face recognition Large-scale identification Watch list Anonymous biometrics
More informationMultimodal Biometric Fusion Joint Typist (Keystroke) and Speaker Verification
Multimodal Biometric Fusion Joint Typist (Keystroke) and Speaker Verification Jugurta R. Montalvão Filho and Eduardo O. Freire Abstract Identity verification through fusion of features from keystroke dynamics
More informationLikelihood Ratio Based Biometric Score Fusion
1 To appear in IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007 Based Biometric Score Fusion Karthik Nandakumar, Student Member, IEEE, Yi Chen, Student Member, IEEE, Sarat C. Dass,
More information8 STOCHASTIC SIMULATION
8 STOCHASTIC SIMULATIO 59 8 STOCHASTIC SIMULATIO Whereas in optimization we seek a set of parameters x to minimize a cost, or to maximize a reward function J( x), here we pose a related but different question.
More informationSpeaker Verification Using Accumulative Vectors with Support Vector Machines
Speaker Verification Using Accumulative Vectors with Support Vector Machines Manuel Aguado Martínez, Gabriel Hernández-Sierra, and José Ramón Calvo de Lara Advanced Technologies Application Center, Havana,
More informationData Mining: Concepts and Techniques. (3 rd ed.) Chapter 8. Chapter 8. Classification: Basic Concepts
Data Mining: Concepts and Techniques (3 rd ed.) Chapter 8 1 Chapter 8. Classification: Basic Concepts Classification: Basic Concepts Decision Tree Induction Bayes Classification Methods Rule-Based Classification
More informationStructurally noise resistant classifier for multi-modal person verification
Pattern Recognition Letters 4 (3) 389 399 www.elsevier.com/locate/patrec Structurally noise resistant classifier for multi-modal person verification Conrad Sanderson a,b, *, Kuldip K. Paliwal b a IDIAP,
More informationINTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH TECHNOLOGY
[Gaurav, 2(1): Jan., 2013] ISSN: 2277-9655 IJESRT INTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH TECHNOLOGY Face Identification & Detection Using Eigenfaces Sachin.S.Gurav *1, K.R.Desai 2 *1
More informationSUPERVISED LEARNING: INTRODUCTION TO CLASSIFICATION
SUPERVISED LEARNING: INTRODUCTION TO CLASSIFICATION 1 Outline Basic terminology Features Training and validation Model selection Error and loss measures Statistical comparison Evaluation measures 2 Terminology
More informationEEL 851: Biometrics. An Overview of Statistical Pattern Recognition EEL 851 1
EEL 851: Biometrics An Overview of Statistical Pattern Recognition EEL 851 1 Outline Introduction Pattern Feature Noise Example Problem Analysis Segmentation Feature Extraction Classification Design Cycle
More informationAutomatic Identity Verification Using Face Images
Automatic Identity Verification Using Face Images Sabry F. Saraya and John F. W. Zaki Computer & Systems Eng. Dept. Faculty of Engineering Mansoura University. Abstract This paper presents two types of
More informationPattern Recognition and Machine Learning
Christopher M. Bishop Pattern Recognition and Machine Learning ÖSpri inger Contents Preface Mathematical notation Contents vii xi xiii 1 Introduction 1 1.1 Example: Polynomial Curve Fitting 4 1.2 Probability
More informationRecent Advances in Bayesian Inference Techniques
Recent Advances in Bayesian Inference Techniques Christopher M. Bishop Microsoft Research, Cambridge, U.K. research.microsoft.com/~cmbishop SIAM Conference on Data Mining, April 2004 Abstract Bayesian
More informationScore Normalization in Multimodal Biometric Systems
Score Normalization in Multimodal Biometric Systems Karthik Nandakumar and Anil K. Jain Michigan State University, East Lansing, MI Arun A. Ross West Virginia University, Morgantown, WV http://biometrics.cse.mse.edu
More informationBAYESIAN DECISION THEORY
Last updated: September 17, 2012 BAYESIAN DECISION THEORY Problems 2 The following problems from the textbook are relevant: 2.1 2.9, 2.11, 2.17 For this week, please at least solve Problem 2.3. We will
More informationCorrespondence. The Effect of Correlation and Performances of Base-Experts on Score Fusion
IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS 1 Correspondence The Effect of Correlation and Performances of Base-Experts on Score Fusion Yishu Liu, Lihua Yang, and Ching Y. Suen Abstract
More informationMaximum Likelihood and Maximum A Posteriori Adaptation for Distributed Speaker Recognition Systems
Maximum Likelihood and Maximum A Posteriori Adaptation for Distributed Speaker Recognition Systems Chin-Hung Sit 1, Man-Wai Mak 1, and Sun-Yuan Kung 2 1 Center for Multimedia Signal Processing Dept. of
More informationSupport Vector Machines using GMM Supervectors for Speaker Verification
1 Support Vector Machines using GMM Supervectors for Speaker Verification W. M. Campbell, D. E. Sturim, D. A. Reynolds MIT Lincoln Laboratory 244 Wood Street Lexington, MA 02420 Corresponding author e-mail:
More informationPATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 2: PROBABILITY DISTRIBUTIONS
PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 2: PROBABILITY DISTRIBUTIONS Parametric Distributions Basic building blocks: Need to determine given Representation: or? Recall Curve Fitting Binary Variables
More informationEstimation of Relative Operating Characteristics of Text Independent Speaker Verification
International Journal of Engineering Science Invention Volume 1 Issue 1 December. 2012 PP.18-23 Estimation of Relative Operating Characteristics of Text Independent Speaker Verification Palivela Hema 1,
More informationHow to Deal with Multiple-Targets in Speaker Identification Systems?
How to Deal with Multiple-Targets in Speaker Identification Systems? Yaniv Zigel and Moshe Wasserblat ICE Systems Ltd., Audio Analysis Group, P.O.B. 690 Ra anana 4307, Israel yanivz@nice.com Abstract In
More informationINTRODUCTION TO PATTERN RECOGNITION
INTRODUCTION TO PATTERN RECOGNITION INSTRUCTOR: WEI DING 1 Pattern Recognition Automatic discovery of regularities in data through the use of computer algorithms With the use of these regularities to take
More informationRobust Speaker Identification
Robust Speaker Identification by Smarajit Bose Interdisciplinary Statistical Research Unit Indian Statistical Institute, Kolkata Joint work with Amita Pal and Ayanendranath Basu Overview } } } } } } }
More informationIf you wish to cite this paper, please use the following reference:
This is an accepted version of a paper published in Proceedings of the st IEEE International Workshop on Information Forensics and Security (WIFS 2009). If you wish to cite this paper, please use the following
More informationIncrease of coal burning efficiency via automatic mathematical modeling. Patrick Bangert algorithmica technologies GmbH 1 Germany
Increase of coal burning efficiency via automatic mathematical modeling Patrick Bangert algorithmica technologies GmbH 1 Germany Abstract The entire process of a coal power plant from coal delivery to
More informationHMM part 1. Dr Philip Jackson
Centre for Vision Speech & Signal Processing University of Surrey, Guildford GU2 7XH. HMM part 1 Dr Philip Jackson Probability fundamentals Markov models State topology diagrams Hidden Markov models -
More informationRevisiting Doddington s Zoo: A Systematic Method to Assess User-dependent Variabilities
Revisiting Doddington s Zoo: A Systematic Method to Assess User-dependent Variabilities, Norman Poh,, Samy Bengio and Arun Ross, IDIAP Research Institute, CP 59, 9 Martigny, Switzerland, Ecole Polytechnique
More informationCombining Accuracy and Prior Sensitivity for Classifier Design Under Prior Uncertainty
Combining Accuracy and Prior Sensitivity for Classifier Design Under Prior Uncertainty Thomas Landgrebe and Robert P.W. Duin Elect. Eng., Maths and Comp. Sc., Delft University of Technology, The Netherlands
More informationSound Recognition in Mixtures
Sound Recognition in Mixtures Juhan Nam, Gautham J. Mysore 2, and Paris Smaragdis 2,3 Center for Computer Research in Music and Acoustics, Stanford University, 2 Advanced Technology Labs, Adobe Systems
More informationThe exam is closed book, closed notes except your one-page (two sides) or two-page (one side) crib sheet.
CS 189 Spring 013 Introduction to Machine Learning Final You have 3 hours for the exam. The exam is closed book, closed notes except your one-page (two sides) or two-page (one side) crib sheet. Please
More informationNon-parametric Methods
Non-parametric Methods Machine Learning Alireza Ghane Non-Parametric Methods Alireza Ghane / Torsten Möller 1 Outline Machine Learning: What, Why, and How? Curve Fitting: (e.g.) Regression and Model Selection
More informationAruna 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 informationISCA Archive
ISCA Archive http://www.isca-speech.org/archive ODYSSEY04 - The Speaker and Language Recognition Workshop Toledo, Spain May 3 - June 3, 2004 Analysis of Multitarget Detection for Speaker and Language Recognition*
More informationModifying Voice Activity Detection in Low SNR by correction factors
Modifying Voice Activity Detection in Low SNR by correction factors H. Farsi, M. A. Mozaffarian, H.Rahmani Department of Electrical Engineering University of Birjand P.O. Box: +98-9775-376 IRAN hfarsi@birjand.ac.ir
More informationMachine Learning for Large-Scale Data Analysis and Decision Making A. Neural Networks Week #6
Machine Learning for Large-Scale Data Analysis and Decision Making 80-629-17A Neural Networks Week #6 Today Neural Networks A. Modeling B. Fitting C. Deep neural networks Today s material is (adapted)
More informationMachine Learning Linear Classification. Prof. Matteo Matteucci
Machine Learning Linear Classification Prof. Matteo Matteucci Recall from the first lecture 2 X R p Regression Y R Continuous Output X R p Y {Ω 0, Ω 1,, Ω K } Classification Discrete Output X R p Y (X)
More informationComparing Robustness of Pairwise and Multiclass Neural-Network Systems for Face Recognition
Comparing Robustness of Pairwise and Multiclass Neural-Network Systems for Face Recognition J. Uglov, V. Schetinin, C. Maple Computing and Information System Department, University of Bedfordshire, Luton,
More informationMidterm. Introduction to Machine Learning. CS 189 Spring Please do not open the exam before you are instructed to do so.
CS 89 Spring 07 Introduction to Machine Learning Midterm Please do not open the exam before you are instructed to do so. The exam is closed book, closed notes except your one-page cheat sheet. Electronic
More informationAn Asynchronous Hidden Markov Model for Audio-Visual Speech Recognition
An Asynchronous Hidden Markov Model for Audio-Visual Speech Recognition Samy Bengio Dalle Molle Institute for Perceptual Artificial Intelligence (IDIAP) CP 592, rue du Simplon 4, 1920 Martigny, Switzerland
More informationMixtures of Gaussians with Sparse Structure
Mixtures of Gaussians with Sparse Structure Costas Boulis 1 Abstract When fitting a mixture of Gaussians to training data there are usually two choices for the type of Gaussians used. Either diagonal or
More informationMachine Learning Ensemble Learning I Hamid R. Rabiee Jafar Muhammadi, Alireza Ghasemi Spring /
Machine Learning Ensemble Learning I Hamid R. Rabiee Jafar Muhammadi, Alireza Ghasemi Spring 2015 http://ce.sharif.edu/courses/93-94/2/ce717-1 / Agenda Combining Classifiers Empirical view Theoretical
More informationCurve Fitting Re-visited, Bishop1.2.5
Curve Fitting Re-visited, Bishop1.2.5 Maximum Likelihood Bishop 1.2.5 Model Likelihood differentiation p(t x, w, β) = Maximum Likelihood N N ( t n y(x n, w), β 1). (1.61) n=1 As we did in the case of the
More informationFeature selection and classifier performance in computer-aided diagnosis: The effect of finite sample size
Feature selection and classifier performance in computer-aided diagnosis: The effect of finite sample size Berkman Sahiner, a) Heang-Ping Chan, Nicholas Petrick, Robert F. Wagner, b) and Lubomir Hadjiiski
More informationTwo-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 informationAutomated Segmentation of Low Light Level Imagery using Poisson MAP- MRF Labelling
Automated Segmentation of Low Light Level Imagery using Poisson MAP- MRF Labelling Abstract An automated unsupervised technique, based upon a Bayesian framework, for the segmentation of low light level
More informationε ε
The 8th International Conference on Computer Vision, July, Vancouver, Canada, Vol., pp. 86{9. Motion Segmentation by Subspace Separation and Model Selection Kenichi Kanatani Department of Information Technology,
More informationCS 543 Page 1 John E. Boon, Jr.
CS 543 Machine Learning Spring 2010 Lecture 05 Evaluating Hypotheses I. Overview A. Given observed accuracy of a hypothesis over a limited sample of data, how well does this estimate its accuracy over
More informationScore calibration for optimal biometric identification
Score calibration for optimal biometric identification (see also NIST IBPC 2010 online proceedings: http://biometrics.nist.gov/ibpc2010) AI/GI/CRV 2010, Ottawa Dmitry O. Gorodnichy Head of Video Surveillance
More informationBrief Introduction of Machine Learning Techniques for Content Analysis
1 Brief Introduction of Machine Learning Techniques for Content Analysis Wei-Ta Chu 2008/11/20 Outline 2 Overview Gaussian Mixture Model (GMM) Hidden Markov Model (HMM) Support Vector Machine (SVM) Overview
More informationWhen enough is enough: early stopping of biometrics error rate testing
When enough is enough: early stopping of biometrics error rate testing Michael E. Schuckers Department of Mathematics, Computer Science and Statistics St. Lawrence University and Center for Identification
More informationMarkov chain Monte Carlo methods for visual tracking
Markov chain Monte Carlo methods for visual tracking Ray Luo rluo@cory.eecs.berkeley.edu Department of Electrical Engineering and Computer Sciences University of California, Berkeley Berkeley, CA 94720
More informationAlgorithm-Independent Learning Issues
Algorithm-Independent Learning Issues Selim Aksoy Department of Computer Engineering Bilkent University saksoy@cs.bilkent.edu.tr CS 551, Spring 2007 c 2007, Selim Aksoy Introduction We have seen many learning
More informationClassification and Pattern Recognition
Classification and Pattern Recognition Léon Bottou NEC Labs America COS 424 2/23/2010 The machine learning mix and match Goals Representation Capacity Control Operational Considerations Computational Considerations
More informationHuman Pose Tracking I: Basics. David Fleet University of Toronto
Human Pose Tracking I: Basics David Fleet University of Toronto CIFAR Summer School, 2009 Looking at People Challenges: Complex pose / motion People have many degrees of freedom, comprising an articulated
More informationAlgorithm Independent Topics Lecture 6
Algorithm Independent Topics Lecture 6 Jason Corso SUNY at Buffalo Feb. 23 2009 J. Corso (SUNY at Buffalo) Algorithm Independent Topics Lecture 6 Feb. 23 2009 1 / 45 Introduction Now that we ve built an
More informationA Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement
A Variance Modeling Framework Based on Variational Autoencoders for Speech Enhancement Simon Leglaive 1 Laurent Girin 1,2 Radu Horaud 1 1: Inria Grenoble Rhône-Alpes 2: Univ. Grenoble Alpes, Grenoble INP,
More informationDesign and Implementation of CUSUM Exceedance Control Charts for Unknown Location
Design and Implementation of CUSUM Exceedance Control Charts for Unknown Location MARIEN A. GRAHAM Department of Statistics University of Pretoria South Africa marien.graham@up.ac.za S. CHAKRABORTI Department
More informationCOM336: Neural Computing
COM336: Neural Computing http://www.dcs.shef.ac.uk/ sjr/com336/ Lecture 2: Density Estimation Steve Renals Department of Computer Science University of Sheffield Sheffield S1 4DP UK email: s.renals@dcs.shef.ac.uk
More informationA 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 informationMODULE -4 BAYEIAN LEARNING
MODULE -4 BAYEIAN LEARNING CONTENT Introduction Bayes theorem Bayes theorem and concept learning Maximum likelihood and Least Squared Error Hypothesis Maximum likelihood Hypotheses for predicting probabilities
More informationLecture 2. Judging the Performance of Classifiers. Nitin R. Patel
Lecture 2 Judging the Performance of Classifiers Nitin R. Patel 1 In this note we will examine the question of how to udge the usefulness of a classifier and how to compare different classifiers. Not only
More informationComparative 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 informationAnalytical Study of Biometrics Normalization and Fusion Techniques For Designing a Multimodal System
Volume Issue 8, November 4, ISSN No.: 348 89 3 Analytical Study of Biometrics Normalization and Fusion Techniques For Designing a Multimodal System Divya Singhal, Ajay Kumar Yadav M.Tech, EC Department,
More informationIntroduction to Probability and Statistics (Continued)
Introduction to Probability and Statistics (Continued) Prof. icholas Zabaras Center for Informatics and Computational Science https://cics.nd.edu/ University of otre Dame otre Dame, Indiana, USA Email:
More informationThe Cost of Dichotomizing Continuous Labels for Binary Classification Problems: Deriving a Bayesian-Optimal Classifier
IEEE TRANSACTIONS ON AFFECTIVE COMPUTING, VOL. XX, NO. X, SEPTEMBER 215 1 The Cost of Dichotomizing Continuous Labels for Binary Classification Problems: Deriving a Bayesian-Optimal Classifier Soroosh
More informationLinear vs Non-linear classifier. CS789: Machine Learning and Neural Network. Introduction
Linear vs Non-linear classifier CS789: Machine Learning and Neural Network Support Vector Machine Jakramate Bootkrajang Department of Computer Science Chiang Mai University Linear classifier is in the
More informationBayesian decision theory Introduction to Pattern Recognition. Lectures 4 and 5: Bayesian decision theory
Bayesian decision theory 8001652 Introduction to Pattern Recognition. Lectures 4 and 5: Bayesian decision theory Jussi Tohka jussi.tohka@tut.fi Institute of Signal Processing Tampere University of Technology
More informationLinear Models for Classification
Linear Models for Classification Oliver Schulte - CMPT 726 Bishop PRML Ch. 4 Classification: Hand-written Digit Recognition CHINE INTELLIGENCE, VOL. 24, NO. 24, APRIL 2002 x i = t i = (0, 0, 0, 1, 0, 0,
More informationIDIAP. Martigny - Valais - Suisse ADJUSTMENT FOR THE COMPENSATION OF MODEL MISMATCH IN SPEAKER VERIFICATION. Frederic BIMBOT + Dominique GENOUD *
R E S E A R C H R E P O R T IDIAP IDIAP Martigny - Valais - Suisse LIKELIHOOD RATIO ADJUSTMENT FOR THE COMPENSATION OF MODEL MISMATCH IN SPEAKER VERIFICATION Frederic BIMBOT + Dominique GENOUD * IDIAP{RR
More informationNonparametric Bayesian Methods (Gaussian Processes)
[70240413 Statistical Machine Learning, Spring, 2015] Nonparametric Bayesian Methods (Gaussian Processes) Jun Zhu dcszj@mail.tsinghua.edu.cn http://bigml.cs.tsinghua.edu.cn/~jun State Key Lab of Intelligent
More informationSINGLE CHANNEL SPEECH MUSIC SEPARATION USING NONNEGATIVE MATRIX FACTORIZATION AND SPECTRAL MASKS. Emad M. Grais and Hakan Erdogan
SINGLE CHANNEL SPEECH MUSIC SEPARATION USING NONNEGATIVE MATRIX FACTORIZATION AND SPECTRAL MASKS Emad M. Grais and Hakan Erdogan Faculty of Engineering and Natural Sciences, Sabanci University, Orhanli
More informationLinear Dependency Between and the Input Noise in -Support Vector Regression
544 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 14, NO. 3, MAY 2003 Linear Dependency Between the Input Noise in -Support Vector Regression James T. Kwok Ivor W. Tsang Abstract In using the -support vector
More informationMultiple Similarities Based Kernel Subspace Learning for Image Classification
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
More informationExperiments with a Gaussian Merging-Splitting Algorithm for HMM Training for Speech Recognition
Experiments with a Gaussian Merging-Splitting Algorithm for HMM Training for Speech Recognition ABSTRACT It is well known that the expectation-maximization (EM) algorithm, commonly used to estimate hidden
More informationModel-based unsupervised segmentation of birdcalls from field recordings
Model-based unsupervised segmentation of birdcalls from field recordings Anshul Thakur School of Computing and Electrical Engineering Indian Institute of Technology Mandi Himachal Pradesh, India Email:
More informationApproximate Inference Part 1 of 2
Approximate Inference Part 1 of 2 Tom Minka Microsoft Research, Cambridge, UK Machine Learning Summer School 2009 http://mlg.eng.cam.ac.uk/mlss09/ Bayesian paradigm Consistent use of probability theory
More informationEngineering Part IIB: Module 4F10 Statistical Pattern Processing Lecture 6: Multi-Layer Perceptrons I
Engineering Part IIB: Module 4F10 Statistical Pattern Processing Lecture 6: Multi-Layer Perceptrons I Phil Woodland: pcw@eng.cam.ac.uk Michaelmas 2012 Engineering Part IIB: Module 4F10 Introduction In
More informationCONDITIONAL MORPHOLOGY FOR IMAGE RESTORATION
CONDTONAL MORPHOLOGY FOR MAGE RESTORATON C.S. Regauoni, E. Stringa, C. Valpreda Department of Biophysical and Electronic Engineering University of Genoa - Via all opera Pia 11 A, Genova taly 16 145 ABSTRACT
More informationFace 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 informationApproximate Inference Part 1 of 2
Approximate Inference Part 1 of 2 Tom Minka Microsoft Research, Cambridge, UK Machine Learning Summer School 2009 http://mlg.eng.cam.ac.uk/mlss09/ 1 Bayesian paradigm Consistent use of probability theory
More informationStatistical and Learning Techniques in Computer Vision Lecture 2: Maximum Likelihood and Bayesian Estimation Jens Rittscher and Chuck Stewart
Statistical and Learning Techniques in Computer Vision Lecture 2: Maximum Likelihood and Bayesian Estimation Jens Rittscher and Chuck Stewart 1 Motivation and Problem In Lecture 1 we briefly saw how histograms
More informationShankar Shivappa University of California, San Diego April 26, CSE 254 Seminar in learning algorithms
Recognition of Visual Speech Elements Using Adaptively Boosted Hidden Markov Models. Say Wei Foo, Yong Lian, Liang Dong. IEEE Transactions on Circuits and Systems for Video Technology, May 2004. Shankar
More informationBiometric Fusion: Does Modeling Correlation Really Matter?
To appear in Proc. of IEEE 3rd Intl. Conf. on Biometrics: Theory, Applications and Systems BTAS 9), Washington DC, September 9 Biometric Fusion: Does Modeling Correlation Really Matter? Karthik Nandakumar,
More informationFusion of Dependent and Independent Biometric Information Sources
Fusion of Dependent and Independent Biometric Information Sources Dongliang Huang ECE, University of Calgary Henry Leung ECE, University of Calgary Winston Li ECE, University of Calgary Department of Electrical
More informationECE 661: Homework 10 Fall 2014
ECE 661: Homework 10 Fall 2014 This homework consists of the following two parts: (1) Face recognition with PCA and LDA for dimensionality reduction and the nearest-neighborhood rule for classification;
More informationScale-Invariance of Support Vector Machines based on the Triangular Kernel. Abstract
Scale-Invariance of Support Vector Machines based on the Triangular Kernel François Fleuret Hichem Sahbi IMEDIA Research Group INRIA Domaine de Voluceau 78150 Le Chesnay, France Abstract This paper focuses
More informationLecture Notes 1: Vector spaces
Optimization-based data analysis Fall 2017 Lecture Notes 1: Vector spaces In this chapter we review certain basic concepts of linear algebra, highlighting their application to signal processing. 1 Vector
More informationA Generative Model Based Kernel for SVM Classification in Multimedia Applications
Appears in Neural Information Processing Systems, Vancouver, Canada, 2003. A Generative Model Based Kernel for SVM Classification in Multimedia Applications Pedro J. Moreno Purdy P. Ho Hewlett-Packard
More informationMixtures of Gaussians with Sparse Regression Matrices. Constantinos Boulis, Jeffrey Bilmes
Mixtures of Gaussians with Sparse Regression Matrices Constantinos Boulis, Jeffrey Bilmes {boulis,bilmes}@ee.washington.edu Dept of EE, University of Washington Seattle WA, 98195-2500 UW Electrical Engineering
More informationA 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