Face Recognition under Real-world Conditions

Ladislav Lenc and Pavel Kr´al

Department of Computer Science and Engineering, University of West Bohemia, Plzeˇn, Czech Republic

Keywords:

Automatic Face Recognition, Czech News Agency, Conﬁdence Measures, Gabor Wavelets, Scale Invariant

Feature Transform.

Abstract:

This paper deals with Automatic Face Recognition (AFR). The main contribution of this work consists in

the evaluation of our two previously proposed AFR methods in real conditions. At ﬁrst, we compare and

evaluate the recognition accuracy of two AFR methods on well-controlled face database. Then we compare

these results with the recognition accuracy on a real-world database of comparable size. For such comparison,

we use a sub-set of the newly created Czech News Agency (

CTK) database. This database is created from

the real photos acquired by the

CTK and the creation of this corpus represents the second contribution of this

work. The experiments show the signiﬁcant differences in the results on the controlled and real-world data.

100% accuracy is achieved on the ORL database while only 72.7% is the best score for the

CTK database.

Further experiments show, how the recognition rate is inﬂuenced by the number of training images for each

person and by the size of the database. We also demonstrate, that the recognition rate decreases signiﬁcantly

with larger database. We propose a conﬁdence measure technique as a solution to identify and to ﬁlter-out the

incorrectly recognized faces. We further show that conﬁdence measure is very beneﬁcial for AFR under real

conditions.

1 INTRODUCTION

Nowadays, Automatic Face Recognition (AFR) has

been one of the most progressive biometric methods.

The spectrum of the possible applications of the AFR

is really broad. It reaches from the surveillance of

people (police), access control to restricted areas to

the labeling of the photos used in the recently very

popular photo sharing applications or in the social

networks.

Many algorithms for the face recognition were

proposed. Most of them perform well under cer-

tain “good” conditions (face images are well aligned,

the same face pose and lighting conditions, etc.).

However, their performance is signiﬁcantly degraded

when these conditions are not accomplished. Many

methods have been introduced to handle these limi-

tations, but only few of them perform satisfactorily

in a fully uncontrolled environment. The main goal

of this paper is thus to evaluate two previously pro-

posed methods (Lenc and Kr´al, 2012a; Lenc and Kr´al,

2012b) under real conditions. The evaluation will be

performed on a huge database owned by the Czech

News Agency (

CTK).

To improve the results of the AFR methods, using

the conﬁdence measures should be beneﬁcial. There-

fore, we further use the previously proposed conﬁ-

dence measure approach (Lenc and Kr´al, 2011) in the

post-processing step. This way we can classify cer-

tain amount of faces with high recognition accuracy

while the rest of faces remains unclassiﬁed.

The results of this work will be used by the

CTK in

order to label the unlabeled photos in the large photo-

database (about 2 millions pictures). Note that only

few labeled images of every person are available. Un-

fortunately, the labeled images don’t contain only the

head/face of the person, but also some other addi-

tional useless information (other people, background

objects, etc.). Therefore, another goal of this work

consists in the proposition of the algorithm for the

corpus creation.

The following section gives an overview of some

important AFR methods. Section 3 describes in more

detail the two AFR methods we used in this work;

namely the adapted Kepenekci method and the SIFT

based Kepenekci approach. Conﬁdence measures are

also described in this section. Section 4 ﬁrst intro-

duces the corpora used for testing. Then, the algo-

rithm for the corpus creation is presented. Further,

there is the description of the experiments performed

to evaluate the methods. In the last section we dis-

cuss the achieved results and give the direction of the

250

Lenc L. and Král P..

Face Recognition under Real-world Conditions.

DOI: 10.5220/0004237402500256

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 250-256

ISBN: 978-989-8565-39-6

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

further research.

2 RELATED WORK

One of the ﬁrst successful approaches is the Prin-

cipal Component Analysis (PCA), so called Eigen-

faces (Turk and Pentland, 1991). It is a statistical

method that takes into account the whole image as

a vector. Eigenvectors of the image matrix are used

for face representation. The method is sensitive on

variations in lighting conditions, pose and scale.

Another method, Fisherfaces, is derived from the

Fisher’s Linear Discriminant (FLD). A description of

this method is presented in (Belhumeur et al., 1997).

Similarly to the Eigenfaces, the Fisherfaces project an

image into another, less dimensional, space. Accord-

ing to the authors, this approach should be insensitive

to changing lighting conditions.

Another group of approaches use Neural Net-

works (NNs). Several NNs topologies were proposed.

One of the best performing methods based on neural

networks is presented in (Lawrence et al., 1997).

Hidden Markov Models (HMMs) are also suc-

cessfully used for AFR (Samaria and Young, 1994).

The method was tested on a dataset containing 5 im-

ages of each of the 24 individuals. Indicated recogni-

tion rate of this approach is 84%. For comparison, the

Eigenfaces were tested using the same dataset and the

recognition rate of 74% is reported.

In (Neﬁan and Hayes, 1998) another HMM-based

approach is described. It is stated there, that the

method signiﬁcantly reduces the computational com-

plexity in comparison with the older methods while

the recognition rate remains the same.

In the last couple of years, several successful

approaches based on Gabor wavelets were intro-

duced (Shen and Bai, 2006). One of the ﬁrst such

methods was proposed by Lades (Lades et al., 1993).

Some approaches (Shen, 2005) also combine the pre-

processing with Gabor wavelets with well established

methods such as Eigenfaces, Fisherfaces, etc. An-

other successful approach described in (Wiskott et al.,

1999; Bolme, 2003) is Elastic Bunch Graph Matching

(EBGM). Kepenekci proposes in (Kepenekci, 2001)

an algorithm that outperforms the classical EBGM.

Moreover, he addresses the main issue of the Elastic

Bunch Graph Matching, manual labeling of the land-

marks.

Recently, also the Scale Invariant Feature Trans-

form (SIFT) (Lowe, 1999) is utilized for face recog-

nition. The algorithm was originally developed for

object recognition. The SIFT features has the ability

to detect and describe local features in images. The

features are invariant to image scaling, translation and

rotation. Moreover, they are also partly invariant to

changes in illumination. When used in the AFR, the

SIFT feature vectors of the reference and test images

are compared using the Euclidean distance.

One of the ﬁrst applications of this algorithm for

the AFR is proposed in (Aly, 2006). It takes the orig-

inal SIFT algorithm and creates the set of descriptors

as described in Section 3. Another approach called

Fixed-key-point-SIFT (FSIFT) is presented in (Krizaj

et al., 2010).

3 METHOD DESCRIPTION

This section details the two AFR methods we used

in this work: the adapted Kepenekci method (Lenc

and Kr´al, 2012a) and the SIFT based Kepenekci ap-

proach (Lenc and Kr´al, 2012b).

3.1 Adapted Kepenekci Method

3.1.1 Gabor Filter

Gabor ﬁlter is a sinusoid modulated with a Gaussian.

A basic form of the two dimensional Gabor ﬁlter is

shown in equation 1.

g(x,y;λ,θ,ψ, σ,γ) = exp(−

´x+ γ

´y

2σ

)cos(2π

´x

+ ψ)

(1)

where ´x = xcosθ+ ysinθ, ´y = −xsinθ+ ycosθ, λ is

the wavelength of the cosine factor, θ represents the

orientation of the ﬁlter and ψ is a phase offset, σ and

γ are parameters of the Gaussian envelope, σ is the

standard deviation of the Gaussian and γ deﬁnes the

ellipticity (aspect ratio) of the function.

The set of 40 Gabor ﬁlters with different frequen-

cies and orientations is used to extract local features.

We consider only the real part of the wavelet response.

3.1.2 Face Representation

The image is convolved with all Gabor ﬁlters from

the ﬁlter bank. 40 wavelet responses R

, where j =

1,...,40, are obtained. Each of these responses is

scanned with a sliding window. Assume a square win-

dow W of size w× w . All possible window positions

within the response are evaluated. The center of the

window, denoted (x

), is considered to be a ﬁdu-

cial point iff:

) = max

(x,y)∈W

(x,y) (2)

FaceRecognitionunderReal-worldConditions

251

) >

wi∗ hi

∑

x=1

∑

y=1

(x,y) (3)

where j = 1,...,40, wi and hi are image width and

height respectively.

The feature vector in point (x,y) is created as follows:

v(x,y) = {x,y, R

(x,y),..., R

(x,y)} (4)

The resulting vector thus contains information

about feature point coordinates and values of Gabor

responses in this point.

3.1.3 Face Comparison

The cosine similarity (Tan et al., 2005) is employed

for vector comparison. The similarity between two

vectors thus takes the values in interval [0,1]. Only

the last 40 positions in the vector are considered.

Let us call T a test image and G a gallery image.

For each feature vector t of the face T we determine

a set of relevant vectors g of the face G. Vector g is

relevant iff:

− x

)

+ (y

− y

)

< distanceThreshold (5)

If no relevant vector to vectort is found, vector t is

excluded from the comparison procedure. However,

the most similar vector (from the relevant vector set)

is used for the face similarity computation. The over-

all similarity of two faces OS is computed as an aver-

age of similarities between each pair of corresponding

vectors as:

T,G

= mean{S(t, g),t ∈ T,g ∈ G} (6)

Then, the face with the most similar vector to each

of the test face vectors is determined. The variable C

says how many times the gallery face G

was the clos-

est to some of the vectors of test face T. The similar-

ity is computed as C

where N

is the total number

of feature vectors in G

. Weighted sum of these two

similarities is used for similarity measure:

T,G

= αOS

T,G

+ β

(7)

The size of the sliding window is very important

for the performance of this method. It determines the

number of ﬁducial points detected and inﬂuences its

accuracy. The higher the window size the less ﬁdu-

cial points are detected. On the other hand, search-

ing larger window needs more computation time. In

the comparison stage, the number of ﬁducial points

determines the time needed. The above mentioned

threshold distanceThreshold also inﬂuences the ac-

curacy and the run-time of this method. The smaller

the value of this threshold is, the less comparisons are

needed and the method works faster.

If more than one training example per person is

used, we create a so called “composed face”. It means

that all vectors extracted from the images of one per-

son are put together and the resulting set of vectors is

used as a gallery face.

3.2 SIFT based Kepenekci Method

SIFT

The SIFT algorithm has basically four steps: ex-

trema detection, removal of key-points with low con-

trast, orientation assignment and descriptor calcula-

tion (Krizaj et al., 2010).

To determine the key-point locations, an image

pyramid with re-sampling between each level is cre-

ated. It ensures the scale invariance. Each pixel is

compared with its neighbors. Neighbors in its level

as well as in the two neighboring (lower and higher)

levels are examined. If the pixel is maximum or min-

imum of all the neighboring pixels, it is considered to

be a potential key-point.

For the resulting set of key-points their stability is

determined. Locations with low contrast and unstable

locations along edges are discarded.

Further, the orientation of each key-point is com-

puted. The computation is based upon gradient orien-

tations in the neighborhood of the pixel. The values

are weighted by the magnitudes of the gradient.

The ﬁnal step is the creation of the descriptors.

The computation involves the 16 × 16 neighborhood

of the pixel. Gradient magnitudes and orientations are

computed in each point of the neighborhood. Their

values are weighted by a Gaussian. For each sub-

region of size 4× 4 (16 regions), the orientation his-

tograms are created. Finally, a vector containing 128

(16× 8) values is created.

The original SIFT algorithm is described in de-

tail in (Lowe, 1999; Lowe, 2004b) and (Krizaj et al.,

2010). An example of implementation of the SIFT

algorithm can be found in (Lowe, 2004a).

Face Comparison

To describe a face we create a set of the SIFT features.

For the face comparison the same matching scheme is

used as in the Kepenekci method 3.1.3. We use also a

composed face (see above) if more than one training

example per person is available.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

252

3.3 Conﬁdence Measure

As in many other works (Lleida and Rose, 1996;

Jiang, 2005), our conﬁdence measure for automatic

face recognitionis an estimate of the a posteriori class

probability.

Let the output of our classiﬁer be P(F|C), where

C is the recognized face class and F represents the

face features. The likelihoods P(F|C) are normalized

to compute the a posteriori class probabilities as fol-

lows:

P(C|F) =

P(F|C).P(C)

∑

I∈F I M

P(F|I).P(I)

(8)

F I M represents the set of all faces and P(C) denotes

the prior probability of the face class C.

The relative conﬁdence value method as already

presented in (Lenc and Kr´al, 2011) is used. This ap-

proach computes the difference between the best hy-

pothesis and the second best one:

P∆ = P(

C|F) − max

C6=

(P(C|F)) (9)

Only the faces with P∆ > T are accepted. This

approach aims at identifying the “dominant” faces

among all the other candidates. T is an acceptance

threshold and its optimal value is adjusted experimen-

tally.

4 EXPERIMENTAL SETUP

4.1 Corpora

ORL Database

The ORL database was created at the AT & T Labora-

tories

. It contains images of 40 individuals. 10 pic-

tures for each person are available. The images con-

tain just the face and have black homogeneous back-

ground. They may vary due to three following factors:

1) time of acquisition; 2) head size and pose; 3) light-

ing conditions. The size of the pictures is 92 × 112

pixels. The further description of this database is

in (Li and Jain, 2005).

Czech News Agency (

CTK) Database

This face data-set has been created using the algo-

rithm described in the next section. We created this

corpus in order to evaluate the AFR methods on real-

world data. All faces are extracted from the photo-

graphic data owned by the

CTK.

Figure 1: Examples of one face from the

CTK face corpus.

The number of photographs containing human

faces reaches 2 millions whereas the count of indi-

viduals is about 20 thousand. Out of this amount we

created a sub-set containing faces of 1065 individu-

als. At least 4 images for each person are available.

The pictures size is 128× 128 pixels and have differ-

ent background. The number of the face examples per

person varies from 1 to 10 images.

Figure 1 shows an example of one person from

this corpus.

This corpus has some important characteristics:

• large number of the different people,

• time span between images of one individual is

more than 20 years,

• pose, tilt and lighting conditions vary,

• photos taken under real conditions,

• difﬁculties for the automatic face recognition.

4.2 Corpus Creation Algorithm

Unfortunately, the labeled part of the

CTK corpus

contains not only the face in the photo (a whole per-

son, more faces, some background objects, etc. avail-

able). Therefore, the pre-processing of the photos

is necessary. In order to minimize the human costs,

an semi-automatic corpus creation algorithm is pro-

posed. It is composed of the following tasks:

1. face detection,

2. eyes detection,

3. head rotation,

4. image resizing and conversion to the grayscale,

5. identiﬁcation and deletion of the incorrectly de-

tected faces,

6. manual veriﬁcation

To detect and to extract the faces we used the

face detector implemented in the OpenCV library

http://opencv.willowgarage.com/wiki/. It uses the

well-known Viola-Jones (Viola and Jones, 2001) face

detection algorithm. Once the faces in the image are

detected, we try to correct the position and rotation

of the head. Using Viola-Jones boosted cascade clas-

siﬁer trained for eye detection we determine the posi-

tion of the eyes. If botheyes are successfullydetected,

we rotate and scale the face so that a horizontal posi-

tion of the eyes is ﬁxed. If the detection of the eyes is

not successful the face is used as it is. Next, we resize

the faces to the size of the 128× 128pixel and convert

FaceRecognitionunderReal-worldConditions

253

Figure 2: Examples of the incorrectly identiﬁed faces by the

OpenCV face detector.

it to the grayscale.

Unfortunately, the face detector identiﬁes also a

certain amount (up to 50 % depending on the qual-

ity of images) of non-face images (false positives er-

rors, see Figure 2). These false detections must be

removed. We used a meta-classiﬁer for this task in or-

der to distinguish two classes: face and non-face. We

chose a neural network of the type Multi-Layer Per-

ceptron (MLP) due to its simplicity and good classiﬁ-

cation results in our previous experiments (Kr´al et al.,

2006). The proposed MLP has three layers contain-

ing 1024 (size of the input face vector), 10 and 2 (face

vs. non-face) nodes respectively. The input vector is

created by scanning the image by window of the size

4×4 pixels. An average intensity value in the window

is computed. To train the MLP, we used a set of man-

ually selected face and non-face (250 faces and 250

non-faces) examples. The resulting corpus has been

checked and corrected manually.

Figure 2 shows examples of false positives found

by the face detector.

4.3 Experiments

The following experiments are proposed and evalu-

ated in order to show:

1. the performance of the methods in the laboratory

conditions,

2. a signiﬁcant degradation of the recognition accu-

racy when the laboratory conditions are replaced

with the real ones,

3. the further degradation when the number of rec-

ognized individuals is increasing and the number

of the training examples is decreasing,

4. that the use of the conﬁdence measure is beneﬁ-

cial in order to ﬁlter out the incorrectly recognized

examples.

Both methods: adapted Kepenekci method (see

Section 3.1) and the SIFT based Kepenekci approach

(see Section 3.2) are evaluated on the two corpora de-

scribed above. The cross-validation procedure is used

to evaluate the approaches in all experiments.

http://www.cl.cam.ac.uk/research/dtg/attarchive/

facedatabase.html

Table 1: Recognition results on the ORL database.

Method Kepenekci SIFT

Training Ex. Recognition rate [%]

9 of 10 99.50 100

8 of 10 99.58 99.86

7 of 10 98.75 100

6 of 10 98.39 99.82

5 of 10 97.42 99.50

4 of 10 96.00 99.17

3 of 10 93.48 97.41

2 of 10 90.63 91.88

1 of 10 78.89 81.52

Accuracy in the Laboratory Conditions

Table 1 shows the recognitionrate of both methods on

the ORL database when different number of training

images (1-9) is used and the rest of images is used for

testing. This table shows that the recognition scores

of both approachesare comparableon the “small” and

“artiﬁcial” corpus. Moreover, the obtained recogni-

tion rate is close to 100% when more examples is

used. This experiment has been performed in order

to show that both approaches perform well in the lab-

oratory conditions on a small corpus.

Table 2: Recognition results on the

CTK database.

Method Kepenekci SIFT

Training Ex. Recognition rate [%]

9 of 10 60.81 72.70

8 of 10 57.66 69.07

7 of 10 53.83 65.20

6 of 10 50.10 62.36

5 of 10 47.12 57.21

4 of 10 42.88 51.17

3 of 10 37.55 44.21

2 of 10 32.09 34.57

1 of 10 24.62 22.37

Accuracy in the Real Conditions with a Small

Number of Individuals

Table 2 displays the results of the methods on a sub-

set of the real

CTK corpus of the comparable size as

the ORL database when the number of training exam-

ples varies from 1 to 9. The sub-set contains 37 indi-

viduals, 10 images for each person. This experiment

shows that the SIFT based Kepenekci approach out-

performs signiﬁcantly the adapted Kepenekci method

(the difference about 11% when 9 training examples

used). Therefore, we use only the SIFT method for

further experiments.

Accuracy in the Real Conditions with a Huge

Number of Individuals

Table 3 details the relation among the recognition ac-

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

254

Table 3: Recognition results of the SIFT approach on the

CTK database using different number of training examples

and the different amount of individuals.

Training Ex. Database size (in-

dividuals #)

Recognition rate

[%]

9 of 10 37 72.70

8 of 9 88 56.70

7 of 8 194 49.94

6 of 7 367 41.92

5 of 6 595 33.51

4 of 5 841 27.90

3 of 4 1065 21.31

curacy, the number of the individuals and the num-

ber of the training examples. This table shows that

the recognition rate decreases signiﬁcantly when the

number of individuals increases and the number of

training images decreases. Therefore, we use the con-

ﬁdence measure (see Section 3.3) to post-process the

recognition results and to ﬁlter out the incorrectly rec-

ognized faces.

Conﬁdence Measure

Two experiments with the conﬁdence measure are re-

alized:

1. The number of individuals in the ﬁrst experiment

is close to 200 people. This number is chosen be-

cause it is sufﬁcient in the current version of our

AFR system for the needs of the

CTK. The num-

ber of training examples is 7, one example is used

for testing.

2. In the second experiment, we evaluate the method

on our whole corpus. The number of individuals

is thus 1065 people (3 training examples).

Figure 3 depicts the recognition rate and the num-

ber of the classiﬁed faces of the ﬁrst experiment (194

different individuals). The recognition accuracy with-

out any conﬁdence measure is only about 50%. This

ﬁgure shows, that the recognition accuracy is close to

100% when more than 20% of images is classiﬁed.

This result is interesting for our application because

the main accent is to the correct classiﬁcation. The

amount of 20% of the correctly recognized faces rep-

resents a signiﬁcant part of the corpus.

Figure 4 plots the recognition rate and the num-

ber of the classiﬁed faces of the second experiment

(1065 different individuals). Without any conﬁdence

measure, the recognition rate is only 21%. This ﬁgure

shows, that the recognition accuracy is close to 80%

when more than 20% of images is classiﬁed. This re-

sult is not the perfect one. However, it is very promis-

ing and one of our future research directions focuses

thus on the proposition of better conﬁdence measure

techniques.

100

0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02

Number of classified faces [in %]

Recognition rate [in %]

Threshold

Classified

Recognition rate

Figure 3: Face recognition rates and the numbers of the

identiﬁed faces when 7 of 8 training examples used (T ∈

[0;1] and 194 individuals are recognized).

100

0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02

Number of classified faces [in %]

Recognition rate [in %]

Threshold

Classified

Recognition rate

Figure 4: Face recognition rates and numbers of the identi-

ﬁed faces when 3 of 4 training examples used (T ∈ [0;1]and

1065 individuals are recognized).

5 CONCLUSIONS AND

PERSPECTIVES

In this paper, we have presented the difﬁculties of

the AFR on the faces acquired in uncontrolled en-

vironment. Two previously proposed AFR meth-

ods (adapted Kepenekci method and the SIFT based

Kepenekci approach) are described and evaluated on

the ORL database (laboratory conditions) and on the

sub-set of the

CTK database (real-world corpus). We

have shown that the accuracy of both methods is

close to 100% on the ORL database. The results on

CTK database show that the SIFT based Kepenekci

method has signiﬁcantly better recognition accuracy.

It reaches 72.7% whereas the adapted Kepenekci

method has recognition rate of 60.8% (9 training ex-

amples are used).

We further analysed the accuracy of the SIFT

based Kepenekci approach if the corpus size increases

and the number of training examples decreases. As

supposed, the recognition rate decreases with increas-

ing number of different recognized individuals. We

have also shown that the number of training examples

FaceRecognitionunderReal-worldConditions

255

inﬂuences the accuracy signiﬁcantly. While 9 train-

ing examples is used, the recognition rate is 72.7%. If

we use only 3 training examples the recognition rate

is only 21.3%.

In the last two experiments we employed the con-

ﬁdence measure to post-process the recognition re-

sults. We compared the results when 7 training ex-

amples and 3 examples are used. The results show

that using conﬁdence measure is very beneﬁcial for

AFR under real-world conditions.

It is obvious that the AFR methods are nowadays

capable to recognize faces perfectly under the condi-

tion: the acquisition of the face images must be con-

trolled. If this condition is not accomplished, the task

is much more difﬁcult. Therefore the perspectives of

the further work on recognition of real-world data lay

more in the detection step than in the recognition it-

self. Further increase of image quality will ensure

much better accuracy of the recognition. Using higher

quality images and utilising the conﬁdence measure

will help to create a reliable recognition system.

ACKNOWLEDGEMENTS

This work has been partly supported by the UWB

grant SGS-2010-028 Advanced Computer and Infor-

mation Systems and by the European Regional De-

velopment Fund (ERDF), project NTIS - New Tech-

nologies for Information Society, European Centre of

Excellence, CZ.1.05/1.1.00/02.0090. We also would

like to thank Czech New Agency (

CTK) for support

and for providing the photographic data.

REFERENCES

Aly, M. (2006). Face recognition using sift features.

CNS/Bi/EE report 186.

Belhumeur, P. N., Hespanha, J. a. P., and Kriegman, D. J.

(1997). Eigenfaces vs. ﬁsherfaces: Recognition using

class speciﬁc linear projection. IEEE Transactions on

Pattern Analysis and Machine Intelligence.

Bolme, D. S. (2003). Elastic Bunch Graph Matching. PhD

thesis, Colorado State University.

Jiang, H. (2005). Conﬁdence measures for speech recog-

nition: a survey. Speech Communication, 45(4):455–

470.

Kepenekci, B. (2001). Face Recognition Using Gabor

Wavelet Transform. PhD thesis, The Middle East

Technical University.

Kr´al, P., Cerisara, C., and Kleˇckov´a, J. (2006). Automatic

Dialog Acts Recognition based on Sent ence Struc-

ture. In ICASSP’06, pages 61–64, Toulouse, France.

Krizaj, J., Struc, V., and Pavesic, N. (2010). Adaptation of

sift features for robust face recognition.

Lades, M., Vorbr¨uggen, J. C., Buhmann, J., Lange, J., and

von der Malsburg, C. (1993). Distortion invariant

object recognition in the dynamic link architecture.

IEEE Transactions On Computers.

Lawrence, S., Giles, S., Tsoi, A., and Back, A. (1997).

Face recognition: A convolutional neural network ap-

proach. IEEE Trans. on Neural Networks.

Lenc, L. and Kr´al, P. (2011). Conﬁdence measure for au-

tomatic face recognition. In International Conference

on Knowledge Discovery and Information Retrieval,

Paris, France.

Lenc, L. and Kr´al, P. (2012a). Gabor wavelets for automatic

face recognition. In 38th International Conference on

Current Trends in Theory and Practice of Computer

Science,

Spindler˚uv Ml´yn, Czech Republic.

Lenc, L. and Kr´al, P. (2012b). Novel matching methods

for automatic face recognition using sift. In 8th AIAI

(Artiﬁcial Intelligence Applications and Innovations)

Confence, Halkidiki, Greece.

Li, S. and Jain, A. (2005). Handbook of face recognition.

Springer-Verlag.

Lleida, E. and Rose, R. C. (1996). Likelihood Ratio Decod-

ing and Conﬁdence Measures for Continuous Speech

Recognition. In ICSLP’96, volume 1, pages 478–481,

Philadelphia, USA.

Lowe, D. (2004a). Software for sift.

Lowe, D. G. (1999). Object recognition from local scale-

invariant features. In International Conference on

Computer Vision.

Lowe, D. G. (2004b). Distinctive image features from scale-

invariant keypoints. International Journal of Com-

puter Vision, 2.

Neﬁan, A. V. and Hayes, M. H. (1998). Hidden markov

models for face recognition. In IEEE International

Conference on Acoustics, Speech, and Signal Process-

ing.

Samaria, F. and Young, S. (1994). Hmm-based architecture

for face identiﬁcation. Image and Vision Computing.

Shen, L. (2005). Recognizing Faces - An Approach Based

on Gabor Wavelets. PhD thesis, University of Not-

tingham.

Shen, L. and Bai, L. (2006). A review on gabor wavelets

for face recognition. Pattern Analysis & Applications.

Tan, P.-N., Steinbach, M., and Kumar, V. (2005). Introduc-

tion to Data Mining. Addison-Wesley.

Turk, M. A. and Pentland, A. P. (1991). Face recognition

using eigenfaces. In IEEE Computer Society Confer-

ence on In Computer Vision and Pattern Recognition.

Computer Vision and Pattern Recognition.

Viola, P. and Jones, M. (2001). Rapid object detection using

a boosted cascade of simple features. In Conference

on Computer Vision and Pattern Recognition.

Wiskott, L., Fellous, J.-M., Kr¨uger, N., and von der Mals-

burg, C. (1999). Face recognition by elastic bunch

graph matching. Intelligent Biometric Techniques in

Fingerprint and Face Recognition.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

256