2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

download 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

of 5

Transcript of 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

  • 8/13/2019 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

    1/5

    Face Recognition by Combining Gabor Wavelets and Nearest Neighbor

    Discriminant Analysis

    Kadir Krta, Onur Dolu, Muhittin Gkmen

    Istanbul Technical University, Computer Engineering Dept.

    34469, Maslak, Istanbul, Turkey

    {kirtac, doluo, gokmen}@itu.edu.tr

    Abstract

    One of the successful approaches in face recognition is

    the Gabor wavelets-based approach. The importanceof the Gabor wavelets lie under the fact that the ker-

    nels are similar to the 2-D receptive field profiles of

    the man visual neurons, offering spatial locality, spatial

    frequency and orientation selectivity. In this work, we

    propose a new combination of a Gabor wavelets-based

    method for illumination and expression invariant face

    recognition. It applies the Nearest Neighbor Discrim-

    inant Analysis to the augmented Gabor feature vectors

    obtained by the Gabor wavelets representation of fa-

    cial images. To make use of all the features provided

    by different Gabor kernels, each kernel output is con-

    catenated to form an augmented Gabor feature vector.The feasibility of the proposed method has been suc-

    cessfully tested on Yale database by giving a compari-

    son with its predecessor, NNDA. The effectiveness of the

    method is shown by a comparative performance study

    against standard face recognition methods such as the

    combination of Gabor and Eigenfaces and the combi-

    nation of the Gabor and Fisherfaces, using a subset of

    the FERET database containing a total of 600 facial

    images of 200 subjects exhibiting both illumination and

    facial expression variations. The achieved recognition

    rate of 98 percent in the FERET test shows the efficiency

    of the proposed method.

    1. Introduction

    Due to the highly informative and discriminative nature

    of the face stimuli, face recognition has been regarded

    as a biometric identification application in computer vi-

    sion community. However, because of the 3-D nature of

    the stimuli, the appearance of a single face can change

    dramatically due to the variations of light, pose and ex-

    pressions. One of the successful approaches for robust

    face recognition is the Gabor wavelets-based approach.

    After the pioneering work of Daugman [1], extending

    1-D Gabor wavelets to 2-D, Gabor wavelets have beenextensively used both in many image processing and

    computer vision applications. The efficiency of the Ga-

    bor wavelets lie under the fact that they exhibit similar-

    ity to the 2-D receptive field profiles of the man visual

    neurons, offering spatial locality, spatial frequency and

    orientation selectivity. Therefore, the Gabor wavelets

    representation of facial images should be robust to illu-

    mination and facial expression variations.

    In [2], Lades et al. introduced the use of Gabor

    wavelets for face recognition using the Dynamic Link

    Architecture. The DLA forms and stores deformable

    model graphs whose vertices are labeled by Gabor jets

    computed from a rectangular subgrid centered over the

    object to be stored. It then applies a flexible graph

    matching between model graphs and the image graph.

    In [3], Wiskott et al. extended the DLA architecture

    to Elastic Bunch Graph Matching. They have utilized

    phase information for accurate node localization and

    introduced the bunch graph data structure which com-

    bined jets of a small set of individuals. In recognition,

    matching takes place between a bunch graph and an im-

    age graph. In [4], Donato et al. utilized Gabor wavelets

    for facial action classification. They achieved the best

    results with the Gabor wavelets approach and with an

    ICA-based scheme. In [5], Liu et al. combined Ga-bor wavelets and Enhanced Fisher Linear Discriminant

    Model for face recognition. They applied a set of Ga-

    bor kernels to whole face image and resulted with aug-

    mented Gabor feature vectors. They then applied fea-

    ture extraction on the augmented Gabor feature vectors

    using EFM.

    In this work, we propose a new combination of a Ga-

    bor wavelets-based method, Gabor+NNDA. It applies

    the Nearest Neighbor Discriminant Analysis [6] to the

    978-1-4244-2881-6/08/$25.00 2008 IEEE

  • 8/13/2019 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

    2/5

    augmented Gabor feature vectors obtained by the Ga-

    bor wavelets representation of facial images. To make

    use of all the features provided by different Gabor ker-

    nels, each kernel output is concatenated to form an aug-

    mented Gabor feature vector. The feasibility of the

    proposed method has been successfully tested on Yale

    database [7] by giving a comparison with its predeces-sor, NNDA. The effectiveness of the proposed method

    is shown by a comparative performance study against

    standard face recognition methods such as the combi-

    nation of Gabor and Eigenfaces and the combination of

    the Gabor and Fisherfaces, using a subset of the FERET

    database containing a total of 600 facial images of 200

    subjects exhibiting both illumination and facial expres-

    sion variations. The achieved recognition rate of 98 per-

    cent in the FERET test shows the efficiency of the pro-

    posed method.

    2. Two-dimensional Gabor wavelets

    Gabor wavelets(kernels) are a set of filters ,, whichare defined as,

    ,(z) =k,2

    2 e

    ( k,2z2

    22 )

    [eik,z e2/2], (1)

    whereand indicates the orientation and scale of thekernels, z= (x,y) denotes the image coordinates,.denotes the norm operator andk, is the wave vector.

    Each kernel is a product of a Gaussian envelope func-

    tion and a complex plane wave.The wave vectork,is defined as,

    k,= kei, (2)

    where k=kmax/f and= /8. kmax is defined

    as maximum frequency, is defined as the width ofthe Gaussian along the x and y axis and f is the spac-

    ing factor between kernels in the frequency domain [2].

    Lades et al. investigated=2, f=

    2 andkmax=/2 yielding with optimal results along with 5 scales, {0, . . . , 4}, and 8 orientations, {0, . . . , 7}. In[8], Shen et al. also discussed tuning the Gabor ker-

    nel parameters and after two experiments they showed

    that 5 scales and 8 orientations yielded with optimal

    recognition performance. The first exponential term in

    the square brackets in (1) indicates the oscillatory part

    while the second exponential term compensates for the

    DC value of the kernel, to make the filter independent

    from the absolute intensity of the image. The kernel, ex-

    hibiting complex response, combines a real part(cosine

    part) and an imaginary part(sine part). The wavelets are

    parameterized byk,, which controls the width of the

    Gaussian window and scale and orientation of the os-

    cillatory part. Theparameter determines the ratio ofwindow width to scale, in other words, the number of

    the oscillations under the envelope function [2].

    2.1. Two-dimensional Gabor wavelets-based

    feature representation

    The 2-D Gabor wavelets representation of an image is

    the convolution of the image with a family of kernels

    ,, where is the orientation and is the spatialscale of the kernel. The convolution of an image I(z)with a Gabor kernel, results withG,(z)which isdefined as,

    G,(z) =I(z) ,, (3)where z = (x,y) and denotes the convolution opera-tion. Thus, the set S={G,(z): {0, . . . , 7},

    {0, . . . , 4

    }}forms the Gabor wavelets representation of

    an imageI(z).

    3. Nearest Neighbor Discriminant Analysis

    Nearest neighbor discriminant analysis(NNDA) is a

    nonparametric feature extraction method which forms

    the between-class and within-class scatter matrices in

    a nonparametric way [6]. Considering a c-class prob-

    lem with classes Ci{i = 1, 2, . . . , c} and training samples{x1,x2, . . . ,xN}, the extra-class and intra-class neighborof a sample xnCi is defined as,

    xEn =argminz z xn,z Ci, (4)

    xIn=argminz

    z xn,z Ci,z =xn. (5)

    The nonparametric extra-class and intra-class distances

    are defined as,

    En =xn xEn, (6)In=xn xIn. (7)

    Using the extra-class and intra-class distances defined

    above, the nonparametric between-class and within-

    class scatter matrices are defined as follows,

    SB=N

    n=1

    wn(En)(

    En)

    T, (8)

    SW=N

    n=1

    wn(In)(

    In)

    T, (9)

    wherewnis defined as,

    wn= In

    In+En . (10)

  • 8/13/2019 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

    3/5

    wn is introduced to emphasize the samples in class

    boundaries and deemphasize the samples in class cen-

    ters. Utilizing the fact that the accuracy of the nearest

    neighbor classification can be directly computed by,

    n=

    En

    2

    In

    2, (11)

    Qiu and Wu [6] reached to the following solution for

    the computation of the projection matrix W,

    W=argmaxW

    tr(WT(SB SW)W). (12)

    Thus, the columns of the projection matrixWare them

    leading eigenvectors ofSB SW, corresponding to themgreatest eigenvalues. Due to the computational ineffi-

    ciency of obtaining extra- and intra-class nearest neigh-

    bors when the data dimensionality is high, first PCA is

    applied to reduce the dimension of the data to N-1(the

    rank of the total scatter matrix) by removing the null

    space of the total scatter matrix.

    To keep the nonparametric extra-class and intra-class

    differences of the high dimensional space consisted

    with the projected extra-class and intra-class differ-

    ences, Qiu and Wu [6] proposed stepwise dimensional-

    ity reduction process. In this scheme, the nonparamet-

    ric between-class and within-class matrices are recom-

    puted each time in the current dimensionality. This pro-

    cess is repeated until reaching the desired dimensional-

    ity.

    They also utilizedk-NN classification criterion in the

    training phase. The nonparametric extra- and intra-class

    differences are rewritten as,

    E =x xE[k/2], (13)

    I =x xI([k/2]+1), (14)where,xI([k/2]+1)is defined as the intra-class([k/2] + 1)-

    th nearest neighbor and xE[k/2] is defined as the [k/2]-

    th extra-class nearest neighbor of the sample x. If the

    distance fromx to xI([k/2]+1) is smaller than the distance

    from x to xE[k/2], x will be classified correctly by k-NN

    classifier.

    4. Face recognition by combining Gabor

    wavelets and Nearest Neighbor Discrim-

    inant Analysis

    After obtaining the feature representation described

    in (3), eachG,(z)is downsampled by a scale factorand normalized by zero-mean and unit variance opera-

    tion. Downsampling is carried out by first smoothing

    G,(z) image using a 5 5 Gaussian window and

    then picking out every (k.) th sample both in x

    and y directions, where 1k( w), k Z and wis the width of the G,(z). The training algorithm ofGabor+NNDA is given in Table 1.

    Table 1: Training algorithm of Gabor+NNDA.

    (1)GivenD dimensional samples{x1,x2, . . . ,xN},d-dimensional discriminant subspace is searched.

    (2)Normalize each sample xi to zero-mean and unit

    variance.

    (3)Apply a set of 40 Gabor kernels(5 scales and 8

    orientations) to each samplexi, resulting with

    Gi,,(z); i {1, . . . ,N}, {0, . . . ,7}, {0, . . . ,4}.

    (4)Downsample each filter output Gi,,(z) with a

    factor of to achieve G()i,,(z), and normalize the

    finalG()i,,(z)to zero-mean and unit variance.

    (5)Concatenate the rows(or columns) of each resul-

    tant G()i,,(z) to form an augmented feature vector

    x()i .

    x()i ={G()i,0,0|G()i,0,1| . . . |G()i,7,4}t.

    (6)Form the final Gabor feature matrix X(), by as-

    sembling eachx()i in columns, side by side.

    X() = {x()1 |x()2 | . . . |x()N}.

    (7)Apply PCA to X() to learn the PCA projection

    matrixTpca.

    Tpca= [1|2| . . . |N1], TpcaN1D.

    (8)Project feature matrixX() with the learned PCA

    model.Ypca=Tt

    pcaX().

    (9)Apply NNDA on Ypcato learn the NNDA projec-

    tion matrix.

    Tnnda= [1|2| . . . |d], TnndadN1.

    (10)Project Gabor+PCA feature matrix Ypcawith the

    learned NNDA model.Y=TtnndaYpca.

  • 8/13/2019 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

    4/5

    In recognition, steps (2) to (5) of the training algorithm

    is applied similarly to a test image y, and y() is ob-

    tained. Then,y() is projected using the projection ma-

    trixTnnda,

    y=Ttnnday(). (15)

    Finally, L2 distance measure is applied to identify ywith the label of the closest feature vector in Ga-

    bor+NNDA feature space,

    L2( y,y) = ( y y)t( y y). (16)

    5. Experimental Results

    We performed the first set of experiments on Yale

    database [7], to compare the proposed method with its

    predecessor, NNDA.

    Yale database contains 165 images of 15 subjects.

    There are 11 images per subject, one for each of the

    following facial expressions or configurations: center-light, w/glasses, happy, left-light, w/no glasses, normal,

    right-light, sad, sleepy, surprised, and wink.

    Scaled images of size 64x64 are used in this exper-

    iment. The downsampling factor of Gabor+NNDA is

    64, i.e., = 64. 5 partitions are formed by randomlyselecting 5 images for training and leaving the remain-

    ing 6 images for testing in each trial. Thestep size of

    NNDA is set to 5. The reduced subspace is of 14 dimen-

    sions. Note that the original dimensionality is 64x64

    = 4,096. While the step size is constant, (k, alpha)tuple is changed as, (k, alpha) : k {1, 3, 5}, alpha{0, 1, 2, 3}.The result is shown in Fig. 2.

    (1,0) (1,1) (1,2) (1,3) (3,0) (3,1) (3,2) (3,3) (5,0) (5,1) (5,2) (5,3)

    0.76

    0.78

    0.8

    0.82

    0.84

    0.86

    0.88

    0.9

    0.92

    (k,alpha)

    recognition

    rate

    NNDA

    Gabor+NNDA

    Figure 1: Relative face recognition performance of

    Gabor+NNDA and NNDA on Yale database. Re-

    duced feature dimension is 14. Step size is set to 5.

    From Figure 1, it can be easily observed that Ga-

    bor+NNDA outperformed NNDA on all (k, alpha) tu-ples. Gabor+NNDA reaches 90 percent accuracy in 14

    feature dimension, along with parametersstep size = 5,

    alpha=1, andk=5.The next experiment is the performance compari-

    son of the two standard methods Gabor+Eigenfaces and

    Gabor+Fisherfaces [5], with the proposed method Ga-

    bor+NNDA, using a subset of the FERET database [9].

    The FERET database contains 1564 sets of images fora total of 14,126 images that includes 1199 individuals

    and 365 duplicate sets of images.

    The experiment includes 600 facial images of 200

    subjects such that each subject has three images of size

    256x384 with 256 gray scale levels. Each of these three

    images shows different variations with the following

    configuration: the first image is in neutral expression,

    the second image shows different expression than the

    first one and the last image contains different illumina-

    tion than first two images. First, the centers of the eyes

    are manually detected, then they are aligned to prede-

    fined locations by rotation and scaling transformations.Finally, face image is cropped to the size of 128x128

    to extract the facial region, which is further normalized

    by zero-mean and unit variance operation. The training

    parameters of Gabor+NNDA is set as, step size = 13,

    alpha=1, andk=1. The result is shown in Figure 2.

    0 20 40 60 80 1000

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

    1

    number of features

    recognitionr

    ate

    Gabor+Eigenfaces

    Gabor+Fisherfaces

    Gabor+NNDA

    Figure 2:Comparative face recognition performance

    of Gabor+Eigenfaces, Gabor+Fisherfaces and Ga-

    bor+NNDA, on the augmented Gabor feature vector

    X

    ()

    , downsampled by a factor of 64, i.e., = 64.

    In the experiment, Gabor+NNDA achieves the highest

    recognition rate by 98 percent in 65 feature dimensions

    while Gabor+Fisherfaces achieves a 92.6 percent accu-

    racy and Gabor+Eigenfaces achieves 40.6 percent accu-

    racy in the same feature dimension. Figure 2 shows the

    overall superior performance of the proposed method

    over Gabor+Eigenfaces and Gabor+Fisherfaces meth-

    ods.

  • 8/13/2019 2008 - Face Recognition by Combining Gabor Wavelets and Nearest Neighbor Discriminant Analysis - In

    5/5

    6. Discussion and Conclusion

    2-D Gabor wavelets are optimally localized in the space

    and frequency domains, and result with illumination,

    expression and pose invariant image features.

    Nearest Neighbor Discriminant Analysis(NNDA) [6]was shown to be an efficient nonparametric feature ex-

    traction tool from the point of view of nearest neighbor

    classification. It does not suffer from the small sam-

    ple size problem and it does not need to estimate any

    parametric distribution because of its nonparametric na-

    ture. Moreover, it does not suffer from the singularity

    of the within-class scatter matrix as no matrix inversion

    is required in eigenvector computation. Due to the non-

    parametric structure and stepwise dimensionality reduc-

    tion process, the training time complexity of NNDA is

    greater than that of Fisherfaces method [7]. However,

    NNDA gives higher performance than Fisherfaces in

    terms of classification accuracy and both methods have

    the same time complexity in recognition phase.

    Gabor+NNDA extracts important discriminant fea-

    tures both utilizing the power of Gabor wavelets and

    NNDA. The efficiency of the approach is shown with

    experiments both in Yale and FERET database. It

    achieved a 98 percent classification accuracy in 65 fea-

    ture dimension, outperforming both standard methods

    such as Gabor+Fisherfaces and Gabor+Eigenfaces [5],

    using a 200 class subset of FERET database exhibiting

    both illumination and expression variations. We also

    evidenced the increase of classification accuracy in the

    Yale test, with increasing k and decreasing alpha val-ues, which also agrees with the theoretical results of

    [6]. The FERET test also showed the favor of stepwise

    dimensionality reduction process, by reaching 98 per-

    cent accuracy after 13 steps. However, the chosen step

    sizes greater than 13 did not increase the classification

    accuracy and did not seem to be effective in terms of

    time complexity. In our experiments, training with 400

    FERET images took 80 seconds for 13 steps by using

    Matlab R14 on our Intel P4 3.2 Ghz configuration.

    It should be noted that no suggestion is given in the

    original NNDA work [6], on how to select parameters

    alpha, k and step size. An optimization scheme such

    as Evolutionary Computing can be suggested to pro-

    vide the optimal parameter selection for training. But

    this would also increase the time complexity of the sys-

    tem. Due to the effectiveness of kernel approaches in

    Gabor wavelets-based face recognition [8], NNDA can

    be extended to a kernel approach and Gabor features

    then can be utilized in kernel-NNDA space. Moreover,

    instead of applying a simple distance measure like L2

    norm in classification, more sophisticated classification

    schemes such as Support Vector Machines or Neural

    Networks can be applied. However, it is no guarantee

    that SVM and Neural Network classifiers would be ef-

    fective in Gabor+NNDA feature space.

    Acknowledgments

    This work has been partially supported by

    TUBITAK(National Science Council of Turkey)

    under the grant National Scholarship Programme

    for MSc. Students and TUBITAK project num-

    bered 104E121. The authors would like to thank

    Fatih Kahraman and Abdulkerim apar for valuable

    discussions.

    References

    [1] J. G. Daugman, Uncertainty relation for resolution

    in space, spatial frequency, and orientation optimized

    by two-dimensional visual cortical filters, J. Opt. Soc.Amer., vol. 2, no. 7, pp. 11601169, 1985.

    [2] M. Lades, J. C. Vorbruggen, J. M. Buhmann, J. M. Lange,

    C. von der Malsburg, R. P. Wrtz, and W. Konen, Dis-

    tortion Invariant Object Recognition in the Dynamic Link

    Architecture, IEEE Trans. Computers, vol. 42, no. 3,

    pp. 300311, 1993.

    [3] L. Wiskott, J.-M. Fellous, N. Krger, and C. von der

    Malsburg, Face Recognition by Elastic Bunch Graph

    Matching, IEEE Trans. on Pattern Analysis and Machine

    Intelligence, vol. 19, no. 7, pp. 775779, 1997.

    [4] G. Donato, M. S. Bartlett, J. C. Hager, P. Ekman, and T. J.

    Sejnowski, Classifying Facial Actions, IEEE Trans.

    Pattern Anal. Mach. Intell., vol. 21, no. 10, pp. 974989,

    1999.

    [5] C. Liu and H. Wechsler, Gabor feature based classifica-

    tion using the enhanced fisher linear discriminant model

    for face recognition, IEEE Transactions on Image Pro-

    cessing, vol. 11, no. 4, pp. 467476, 2002.

    [6] X. Qiu and L. Wu, Nearest Neighbor Discriminant Anal-

    ysis,IJPRAI, vol. 20, no. 8, pp. 12451260, 2006.

    [7] P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman,

    Eigenfaces vs. Fisherfaces: Recognition Using Class

    Specific Linear Projection, IEEE Trans. Pattern Anal.

    Mach. Intell., vol. 19, no. 7, pp. 711720, 1997.

    [8] L. Shen, L. Bai, and M. C. Fairhurst, Gabor wavelets

    and General Discriminant Analysis for face identification

    and verification, Image Vision Comput., vol. 25, no. 5,pp. 553563, 2007.

    [9] P. J. Phillips, H. Wechsler, J. Huang, and P. J. Rauss,

    The FERET database and evaluation procedure for face-

    recognition algorithms, Image Vision Comput., vol. 16,

    no. 5, pp. 295306, 1998.