=Paper=
{{Paper
|id=Vol-2267/400-404-paper-76
|storemode=property
|title=Development of software for face retrieval systems modeling
|pdfUrl=https://ceur-ws.org/Vol-2267/400-404-paper-76.pdf
|volume=Vol-2267
|authors=Nadezhda Shchegoleva,Varvara Petrova
}}
==Development of software for face retrieval systems modeling==
https://ceur-ws.org/Vol-2267/400-404-paper-76.pdf
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
DEVELOPMENT OF SOFTWARE FOR FACE RETRIEVAL
SYSTEMS MODELING
N. Shchegoleva a, V. Petrova
Saint Petersburg Electrotechnical University "LETI", ul. Professora Popova 5, St. Petersburg,
197376, Russia
E-mail: a nlschegoleva@etu.ru
The development of software for face retrieval systems modeling is studied. An overview of the state
of the problem is provided. Computer modeling is shown to be required to select the most appropriate
system structure, set of modules and their parameters. The basic requirements for modern face
retrieval systems are determined. It is found that they provided the concept of building a software
complex for FaReS modeling, which formed the basis for a new Simulink library developed by the
authors. Examples of solving practical problems of facial biometrics, structure, composition and
parameters of blocks of used systems are shown. Compact models of computer experiments are
presented.
Keywords: object-oriented modeling, Simulink, library for modeling face recognition systems, model
of compact description of computer experiment
© 2018 Nadezhda Shchegoleva, Varvara Petrova
400
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
1. Introduction
Over the past two decades powerful and affordable computing equipment for solving applied
problems of retrieval and identifying people has been created. The performance of modern equipment
is commensurate with the amount of data in the image database.
The human face, being its unique characteristic, is often used for identification purposes, for
example, in the access control systems of different levels (from the entrance to the laboratory to cross-
border control). In practice, facial images can be represented in various categories - in visible and /or
infrared light, identikits, in the form of "range image" and 3D objects.
On the one hand, the presence of such categories significantly expands the field of application
of facial biometrics, and on the other hand it substantially complicates the solution of applied
problems.
Currently, the main studies are aimed at improving the methods of image processing, but very
little attention is paid to issues related to the development and optimization of systems. This leads to
the fact that a more accurate solution of the problem is provided by increasing the complexity of the
system. The development of effective FaReS (Face Retrieval Systems) is a non-trivial task [1, 2] that
requires not only a good knowledge of facial image processing methods, but also a lot of experience in
designing and programming such systems. To solve this problem, it is advisable to use computer
modeling. However, at the moment, there are practically no full software tools for modeling facial
image processing systems (figure 1).
Figure 1. Methodical support of facial biometrics tasks
2. Concept of building software for modeling FaReS
One of the most important and complex stages is the formulation of the problem and the
choice of an adequate solution to it. However, this requires either a great practical experience or a lot
of experiments, so in practice, preference is often given to systems using universal but excessively
complex and resource-intensive methods for image processing. To search for more economical and
simple solutions, it is reasonable to use computer modeling at the stage of developing the concept of a
future system.
One of the reasons for the lack of FaReS modeling tools is the lack of requirements for such
software. The software for modeling should have a modular structure that will provide flexibility and
extensibility of the system without a significant increase in labor costs for implementation. The
software system should provide the ability to solve the task of each block by several methods, as well
as adding new functional modules, including those developed by users. It is also necessary to
synthesize a system that has a cascading, parallel or more flexible hybrid structure, it is necessary to
401
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
ensure the possibility of using different information about the same person. For the implementation it
is necessary to have the formation of a ready-made application for solving a specific problem,
evaluation of time and computing costs, evaluation of the quality of the developed system. For
research, analysis of the composition, characteristics of the feature space, data randomization,
selection of optimal parameters, experimental planning, visualization of data and results, flexible
reporting. A detailed list of requirements is presented on the figure 2.
Figure 2. Concept of building software for modeling FaReS
3. Simulink library FaReS design
Three groups of modern tools are most often used for modeling: software development
environments, graphic environments for simulation of FaReS, specialized software for modeling
recognition systems. A key advantage for solving the problems of the facial biometry with the help of
Graphical environments of simulation is the Ready-made graphical modeling environment, and for
Simulink also the support of the MATLAB language [3], which allows the processing of images in
terms of matrix algebra.
To solve any applied task of biometrics it is necessary to determine the architecture of the
target system. To do this, you should take into account the quality of the data, the number of data
sets, the dimensionality of the data, the dynamics of data changes, the purpose of the analysis. You
should chose method of feature extraction/selection and feature space dimension reduction methods,
and all of this defines the type of system structure.
As illustrated above, Simulink is the most appropriate option to design a simulation system,
for this you are able to use blocks from standard packages but they are not enough for a full-scale
simulation. Thus, it is necessary to develop new custom blocks.
To FaReS design the authors developed a Simulink library. The steps required for block
creation are open the standard Simulink block library, select the Matlab function block, define the
block interface, describe block behavior in Matlab language, set the sizes of input and output data
streams, add a new block to the developed library.
Figure 3 shows an example of a system developed using this developed library. Each block
has several parameters for tuning, setting values to which you can improve the results of the
system.The parameters can be external to the block (connected as a separate block) or internal
(available in the block settings). The library allows creating cascade or parallel structure (Figures 4)
that provides an ability to generate systems of any complexity.
402
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
Figure 3. Example of setting parameters and workflow of facial image processing system
Figure 4. Example of parallel facial image processing systems
4. Model of compact description of computer experiment
For a compact description of a computer experiment, it is proposed to use a model that more
clearly and fully describes the structure of FaReS, has the ability to describe parallel, cascade and
combined systems [4]. A set of parameters included in the model is sufficient to perform a similar
experiment by other researchers or other platforms, and also for comparative analysis during testing of
simulated systems.
We define the elementary system as a combination of the image base, the block of feature
extraction/selection and the classifier.
ES = DB ( K/L/(Q – L)MOD: TV ) {PO/TF:M×Nd1×d2/SF}[Cls/Met/rank].
The base of images is described by the following parameters: K – number of classes in a
database; Q – number of images in a class; L – number of etalons representing the class; MOD – a
sign of modification; TV – type of modification (cross-validation, randomization).
403
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
The feature extraction/selection is described by the following parameters: M – highest of facial
image; N – width of facial image; TF – representation of facial image: d1×d2 – dimension; PO –
preprocessing; FS – feature selection.
The classifier is described by the following parameters: Cls – classifier type; Met –metrics; rank
– rank of result.
The facial image processing systems description based on compact description shown in Figure
3 will be described as follows
ES ORL (40/4/6) Hist : 96 112 1 64[CMD/L2/rank = 1] .
The description of a simple system fits just one line, it is clear that for a parallel, cascade and
more complex system structure it was also possible to give a brief definition.
The model of cascade system will be
CS = ES 1 ES 2 … ES KS.
Then the model of parallel system will be
ES1
ES
2
PS : Fuz ES Itog ,
KS
ES
where KS – number of basic systems, Fuz – fusion (method of integration) solutions, ESItog – system
forming the solution.
The facial image processing systems description based on compact description Figure 4 will be
described as follows
ORL (40/4/6) Hist : 96 112 1 64[CMD/L2/ra nk = 1]
ORL (40/4/6) Scale : 96 112 12 14[CMD/L2/ra nk = 1]
PS ORL (40/4/6) Rand : 96 112 1 50[CMD/L2/ra nk = 1] : Weighted Voting/ran k 1
ORL (40/4/6) DCT : 96 112 1 210[CMD/L2/ra nk = 1]
ORL (40/4/6) DFT : 96 112 1 210[CMD/L2/ra nk = 1]
5. Conclusion
A methodology for developing FaReS using a graphical library is proposed. It allows
modeling facial image processing systems depending on the scenario of the problem being solved, the
structure of the initial data, the number of data sets, the dynamics of their changes and the chosen
criterion (minimum approximation error, improvement of data clustering, maximum correlation of
variables in the subspace, etc.). The approach allows solving a rather wide range of facial image
processing tasks and creating systems that differ in the smaller number of modules and computational
complexity.
References
[1] Kukharev G., Shchegoleva N. Face recognition systems[In Russian]. Publisher SPbGETU
«LETI»- LETI, St. Petersburg. 2006. – 176 p. ISBN: 5-7629-0665-5.
[2] Kukharev G., Kamenskaya E., Matveev Y., Shchegoleva N. Methods of facial images processing
and recognition in biometrics [In Russian]. Saint-Peterburg: “Politechnika” Publisher; 2013. – 388 p.
ISBN: 978-5-73251-028-7.
[3] Varvara A. Petrova, Nadezhda L. Shchegoleva Rapid Prototyping of Face Retrieval Systems in
Simulink. Proc. XХ International Conference on Soft Computing and Measurements (SCM). 2017. –
P. 312-314.
[4] Shchegoleva N. L. Cconception of software for face retrieval systems modeling. Saint Petersburg
Electrotechnical University «LETI», 2016. – №5 – P. 40-47.
404