Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance
The proposed concept suggests a method, based on which a synthesis of an integer norm can be performed, which takes into account the white balance of the camera when performing the evaluation of the difference between the elements of an image. This idea is based on modifying the internal calculation...
Saved in:
| Date: | 2019 |
|---|---|
| Main Authors: | , , |
| Format: | Article |
| Language: | English |
| Published: |
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН та МОН України
2019
|
| Series: | Управляющие системы и машины |
| Subjects: | |
| Online Access: | https://nasplib.isofts.kiev.ua/handle/123456789/161658 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Cite this: | Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance / M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko // Управляющие системы и машины. — 2019. — № 4. — С. 27-34. — Бібліогр.: 7 назв. — англ. |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
nasplib_isofts_kiev_ua-123456789-161658 |
|---|---|
| record_format |
dspace |
| spelling |
nasplib_isofts_kiev_ua-123456789-1616582025-02-10T01:16:12Z Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance Концепція цілочислової норми оцінки різниці між елементами зображення з врахуванням балансу білого Концепция целочисленной нормы оценки разницы между элементами изображения с учетом баланса белого Ivashchenko, M.V. Okhrymchuk, D.D. Lyushenko, L.A. Intelligent Information Technologies and Systems The proposed concept suggests a method, based on which a synthesis of an integer norm can be performed, which takes into account the white balance of the camera when performing the evaluation of the difference between the elements of an image. This idea is based on modifying the internal calculations of the camera, aimed at assessing the colour of the image element, using the process of colour model reduction that is embedded inside the camera, to the colour model of the classical representation. The use of this approach provides a number of advantages within the framework of systems in which there is a solution of computer vision problems in terms of using both graphical processing and artificial intelligence. Мета. Метою даної статті є дослідження запропонованого підходу, на основі якого можливо виконати синтез цілочислової норми, що враховує баланс білого камери при розрахуванні оцінки різниці між елементами зображення. Методи. Камера автоматично налаштовує баланс білого відповідно до одного з режимів, встановлених за замовчуванням. Оцінка різниці між двома точками зображення здійснюється за рахунок виконання нормалізації. Як метод розрахунку використовується евклідова норма. Результати. Запропоновано підхід до нормалізації, що враховує коефіцієнти корекції балансу білого. Розрахунки виконуються з точки зору здійснення операцій з цілими значеннями, що надає можливість їх використання для безпосереднього розгортання всередині апаратної логіки пристроїв. Цей факт дозволяє знизити як загальну вартість ресурсів для виконуваних операцій, по відношенню до існуючої арифметики з плаваючою крапкою, так і зниження вимог до конфігурації датчиків. Цель. Целью данной статьи является исследование подхода к синтезу целочисленной нормы, которая учитывает баланс белого камеры при расчете оценки разницы между элементами изображения. Методы. Камера автоматически настраивает баланс белого в соответствии с одним из режимов, установленных по умолчанию. Оценка разницы между двумя точками изображения осуществляется путем выполнения нормализации. В качестве метода расчета используется евклидова норма. Результаты. Предложен подход нормализации, учитывающий коэффициенты коррекции баланса белого. Расчеты выполняются исходя из операций с целыми значениями, что дает возможность использования их для непосредственного развертывания внутри аппаратной логики устройств. Это позволяет снизить как общую стоимость ресурсов для выполняемых операций, по отношению к существующей арифметике с плавающей точкой, так и упростить требования к конфигурации датчиков. 2019 Article Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance / M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko // Управляющие системы и машины. — 2019. — № 4. — С. 27-34. — Бібліогр.: 7 назв. — англ. 0130-5395 DOI: https://doi.org/10.15407/csc.2019.04.027 https://nasplib.isofts.kiev.ua/handle/123456789/161658 004.4 451 en Управляющие системы и машины application/pdf Міжнародний науково-навчальний центр інформаційних технологій і систем НАН та МОН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| topic |
Intelligent Information Technologies and Systems Intelligent Information Technologies and Systems |
| spellingShingle |
Intelligent Information Technologies and Systems Intelligent Information Technologies and Systems Ivashchenko, M.V. Okhrymchuk, D.D. Lyushenko, L.A. Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance Управляющие системы и машины |
| description |
The proposed concept suggests a method, based on which a synthesis of an integer norm can be performed, which takes into account the white balance of the camera when performing the evaluation of the difference between the elements of an image. This idea is based on modifying the internal calculations of the camera, aimed at assessing the colour of the image element, using the process of colour model reduction that is embedded inside the camera, to the colour model of the classical representation. The use of this approach provides a number of advantages within the framework of systems in which there is a solution of computer vision problems in terms of using both graphical processing and artificial intelligence. |
| format |
Article |
| author |
Ivashchenko, M.V. Okhrymchuk, D.D. Lyushenko, L.A. |
| author_facet |
Ivashchenko, M.V. Okhrymchuk, D.D. Lyushenko, L.A. |
| author_sort |
Ivashchenko, M.V. |
| title |
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance |
| title_short |
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance |
| title_full |
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance |
| title_fullStr |
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance |
| title_full_unstemmed |
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance |
| title_sort |
integer norm for difference assessment of the frame elements considering the white balance |
| publisher |
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН та МОН України |
| publishDate |
2019 |
| topic_facet |
Intelligent Information Technologies and Systems |
| url |
https://nasplib.isofts.kiev.ua/handle/123456789/161658 |
| citation_txt |
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance / M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko // Управляющие системы и машины. — 2019. — № 4. — С. 27-34. — Бібліогр.: 7 назв. — англ. |
| series |
Управляющие системы и машины |
| work_keys_str_mv |
AT ivashchenkomv integernormfordifferenceassessmentoftheframeelementsconsideringthewhitebalance AT okhrymchukdd integernormfordifferenceassessmentoftheframeelementsconsideringthewhitebalance AT lyushenkola integernormfordifferenceassessmentoftheframeelementsconsideringthewhitebalance AT ivashchenkomv koncepcíâcíločislovoínormiocínkiríznicímíželementamizobražennâzvrahuvannâmbalansubílogo AT okhrymchukdd koncepcíâcíločislovoínormiocínkiríznicímíželementamizobražennâzvrahuvannâmbalansubílogo AT lyushenkola koncepcíâcíločislovoínormiocínkiríznicímíželementamizobražennâzvrahuvannâmbalansubílogo AT ivashchenkomv koncepciâceločislennoinormyocenkiraznicymežduélementamiizobraženiâsučetombalansabelogo AT okhrymchukdd koncepciâceločislennoinormyocenkiraznicymežduélementamiizobraženiâsučetombalansabelogo AT lyushenkola koncepciâceločislennoinormyocenkiraznicymežduélementamiizobraženiâsučetombalansabelogo |
| first_indexed |
2025-12-02T10:30:45Z |
| last_indexed |
2025-12-02T10:30:45Z |
| _version_ |
1850392125526507520 |
| fulltext |
ISSN 2706-8145, Control Systems and Computers, 2019, № 4 27
DOI https://doi.org/10.15407/usim.2019.04.027
UDC 004.4 451
M.V. IVASHCHENKO, Student of the Faculty of Applied Mathematics,
National Technical University of Ukraine "Igor Sikorsky Kyiv PolitechnicI nstitute",
03056, Peremohy Ave, 37, Kyiv, Ukraine,
mivaschenko_51@lll.kpi.ua
D.D. OKHRYMCHUK, Student of the Faculty of Applied Mathematics,
National Technical University of Ukraine "Igor Sikorsky Kyiv PolitechnicI nstitute",
03056, Peremohy Ave, 37, Kyiv, Ukraine,
den5096@gmail.com
L.A. LYUSHENKO, PhD (Eng.), Senior Lecturer,
National Technical University of Ukraine "Igor Sikorsky Kyiv Politechnic Institute",
03056, Peremohy Ave, 37, Kyiv, Ukraine,
lyushenkol@gmail. com
INTEGER NORM FOR DIFFERENCE
ASSESSMENT OF THE FRAME ELEMENTS
CONSIDERING THE WHITE BALANCE
The proposed concept suggests a method, based on which a synthesis of an integer norm can be performed, which takes into ac-
count the white balance of the camera when performing the evaluation of the difference between the elements of an image. This
idea is based on modifying the internal calculations of the camera, aimed at assessing the colour of the element of the image, using
the process of colour model reduction that is embedded inside the camera, to the colour model of the classical representation. The
use of this approach provides a number of advantages within the framework of systems in which there is a solution of computer
vision problems in terms of using both graphical processing and artificial intelligence.
Keywords: computer vision, white balance correction, colour sensors, colour model, reference colours, image point norm.
Introduction
Human organism possesses the ability to see, de-
termine and understand the environment using the
visual information. The same idea was also formed
and implemented in the computer vision technolo-
gies. Such technologies are usually used in surveil-
lance systems, robotics, tracking mobile applica-
tions, engineering reliability and quality control
etc. In every computer vision system, it is needed
to track how the image has changed (or how the
points of the image have changed).
Computer vision systemsbuilt over modern CCD
CMOS-sensors, allow adjust automatically the pa-
rameters of a frame according to the surrounding
illumination, its position changesin space, integral
characteristics of the frame, and other parameters
that can indirectly influence the raster quality.
The increase of the computational power in digi-
tal signal processorsbuilt in the sensor systems of
computer vision allows on the one hand to signifi-
cantly improve consumer and metrological charac-
teristics of the finished device without tightening
the technological requirements, but on the other
hand requires specific processing algorithms that
are attached to the architecture and scheme pa-
rameters of the specific device.
In this paper are presented the results of the re-
search based on the task of forming the difference
assessment between two elements of a raster frame,
that allows the conversion of the incoming data
M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko
28 ISSN 2706-8145, Системи керування та комп'ютери, 2019, № 4
and the results of the comparison on hypothetically
arbitrary bit grids distributed to a set of individual
raster points representation ways, including colour
schemes and device configuration parameters.
Nowadays computer vision is developing in
a wide variety of fields, one of which is machine
learning, specifically deep learning that manages
the high-level data composed into the hierarchi-
cal structures. It improves the abilities of chip-
programming using relatively cheap computational
equipment [1].
A distinctive feature of computer vision engi-
neering and technology is the usage of adaptive al-
gorithms and learning systems that are the basis for
the practical computer vision usage in any valuable
automation tasks.
It leads to active implementation and develop-
ment of the deep learning algorithms, weak artificial
intelligence in terms of configuration and learning
process, artificial intelligence for the recognition
and classification tasks. The mentioned techniques
that are used for solving the issues of signals digi-
tal processing in machine vision allow the active
usage of phase-frequency and adaptive filtering,
spectral analysis methods, wavelet transformation
in the early stages of reading information from the
matrix photo detectors on the programming and
hardware levels. It also allows to reduce signifi-
cantly the computational costs of image processing
algorithms and distribute the processing between
all links of machine vision chains.
Nevertheless, all the tasks described above are re-
duced to the basic task of computer vision, namely:
comparison of image points or series of images.
This process is implemented in terms of work with
the colour models. Each colour model defines the
way of a point’s colour representation and the way
of its encoding inside a device.
The difference assessment between two elements
of a frame is a very important block of all image
processing algorithms. However, nowadays the
most of the norms that are defined over the colour
spaces, perform the operations with the float va-
lues. When implementing algorithms for calcula-
ting the norm into the hardware logic of sensors,
the usage of float arithmetic is unacceptable as it
significantly increases the computational costs and
also the hardware requirements of the sensors cau-
sing a great reduce in the performance level of the
norm computational algorithms in terms of impos-
sibility of their paralleling.
Thus, the development of integer norms over
the colour spaces will allow to organize an efficient
ima-ge processing using the means of a device.
In this article RGB colour model is reviewed.
This model applies the additive method of colour
processing (a colour is obtained by adding to black).
It is known that any colour can be represented as a
combination of three colours: red (R), white (W),
and blue (B). These colours are also known as the
reference colours and are specified by values from
0 to 255. RGB colour model allows the usage of
ima-ge points colour identification of implement-
ing the three-channel colour representation. It al-
lows to perform the image normalization process
exclu-ding the set of operations that are specific for
and are defined over the float arithmetic.
Let us review the described approach.
Problem Description
Let there be a raster rectangular image of the reso-
lution X X W H C, where W is the number of pix-
els in the width of the image, H — the number of
pixels in the height of the image, C — the colour
depth of the image (the number of bits used for
contacting the representation of a colour in terms
of a pixel encoding process). The position of every
point of the image is uniquely determined by a pair
of coordinates ( ); ,X Y where 1.. , 1.. .X W Y H= = In
this case the content of every point is represented
by a colour that is encoded by the device.
Let there be a point of animage that is determined
by a pair ( )1 1; ,X Y where 1 11.. , 1.. .X W Y H= = It is
needed to build an assessment, how this point dif-
fers from another point of the same image with a pair
of coordinates ( )2 2; ,X Y where 2 21.. , 1.. .X W Y H= =
This task can be solved by finding the distance bet-
ween the points on the colour surface.
The content of a point of an image can be given
as a linear combination of a colour model reference
points that is assigned by a polygon on the colour
surface. Thus, in general the difference between the
two colors is to be defined according to the distance
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance
ISSN 2706-8145, Control Systems and Computers, 2019, № 4 29
between the colours. This distance can be comput-
ed as the length of a segment which endpoints are
the positions of the colours on the surface.
Let us review the situation where each of the de-
vices makes certain adjustments to the immediate
appearance of the colour model when performing
white balance correction by deforming it. As a re-
sult, both the appearance of the polygon that limits
the area of the colour plane and determines the in-
ternal colour scheme of the device, and the appea-
rance of the surface that describes the colour gamut
can change. While image processing algorithms
take into account only some standards [2] or use
classical representations of colour schemes, mat-
ching them with the results obtained from the ca-
mera, it makes sense to perform the final distance
calculations considering the changes of the colour
model due to the camera's white balance correc-
tions before obtaining every next frame. As a result,
the colour channel component values will be con-
verted according to the white balance factors cal-
culated after shooting the scene, and subsequently
represented in traditional RGB color model.
Then the final task of finding the difference bet-
ween the two points of an image can be decreased
to the following actions:
performing the reduction of colour values ob-
tained from the device to the values of the original
color model;
computing the distance between the points by
calculating the norm (the problem can be set as a
synthesis of a different type of a norm in terms of
working with integer values and consideration of
the white balance corrections).
Speaking of the problem of the device colour
scheme structure, the solution is to use the refe-
rence colours system conversion and the resulting
change in the appearance of the colour surface, as
well as the color gamut. This step can be performed
as a precomputed set of operations.
Related Works. Colour Conversion
The process of manufacturing, testing and auto-
mated adjustment of sensors involves the rejection
and calibration of individual light-sensitive ele-
ments. The result of reading the image with some
accuracy provides a representation of colours in
a vector form, taking into account the calibrating
parameters.
Let us review the models of obtaining the image
element data according to the results of measure-
ments in individual photosensors.
1. Matrix Method
This method is used if the assessment of several co-
lours is needed. The method is based on the matrix
equation: the values ( ); ;X Y Z represent the colours
in the source colour model, the values ( ); ;R G B are
the representation of the digital data of the colour
sensor. The conversion matrix C is based on the
output signals of the reference colour obtained
during the calibration process. After the matrix
coefficients are defined the values ( ); ;X Y Z can
be computed from the conversion of the ( ); ;R G B
sensor values.
(1)
2. Table Method
This method is used if it is needed to define several
colours simultaneously. The first step is to determine
how important the brightness is. For each colour
the sensor uses brightness information that was ob-
tained during calibration from the reference colours
set. If the brightness information is not important,
then a specifically selected colour channel is used
for the search of the ratios between the reference co-
lors obtained during calibration and the determined
colour (taken as the basis for all measurement sets).
The brightness matters:
(2)
The bright ness does not matter:
2 2
,
R BR Bu ur r
G G G Gu r u r
⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟− − −
⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠
(3)
where ( )1 1 1, , R G B are the coordinates values of the
investigated colour, ( )0 0 0, , R G B — the coordinates
values of the reference colour.
The colours formed by the above methods lead
to an incorrect result in terms of light conditions,
( ) ( ) ( )2 2 2
1 0 1 0 1 0 .R R G G B B− ⋅ − ⋅ −
11 12 13
21 22 23
31 32 33
C C CX R
Y C C C G
Z BC C C
⎛ ⎞⎛ ⎞ ⎛ ⎞
⎜ ⎟⎜ ⎟ ⎜ ⎟= ⋅
⎜ ⎟ ⎜ ⎟⎜ ⎟⎝ ⎠ ⎝ ⎠⎝ ⎠
M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko
30 ISSN 2706-8145, Системи керування та комп'ютери, 2019, № 4
the sensor temperature and the sampling frequency
changes. Attempting to adjust the calibration data
contributes to the accumulation of errors and pro-
cessing algorithms complication [3].
Related Works. White Balance
Correction Algorithms
There are different approaches of white balance
correction during the shooting of a scene (or sce-
nes) using a device. The following section describes
the basic versions of such algorithms.
1. Gray World Theory
This method provides automatic white balance cor-
rection by calculating the values ( ); ;avg avg avgBR G ,
such as ,avg avg avgBR G= = and is based on the fact
that any scene can be converted into the gray chan-
nel. The coefficients are used for colour correction
of the image points are obtained from the calcula-
tions of the ratios between the ( ); ;avg avg avgBR G va-
lues and the average value of the green channel.
,avg avg avgBR G= = max max , corrR G R corrB= =
max max , 1.G B corrG= = (4)
2. Retinex Theory
Another possible method was produced by Edwin
Land [4].This method is based on the principle that
the white colour is a combination of three maximum
values of the colour channels as it transmits the ma-
ximum possible level of signals into every channel.
This method forms the white colour according to
the component values of the brightest point of the
image. At the same time, white balance correc-
tion factors can be obtained in two different ways:
max max max max, , 1 , corr R G R corr B G B corr G= = =
max max, , corr R R R corr B B B= =
max .corr G G G= (5)
3. Standard Deviation-Weighted Gray World
Standard Deviation-Weighted Gray World method
extends the representation of the Gray world
method and was suggested by Lam [5].The ima-
ge is divided into a certain number of blocks. For
each block the standard deviations and values
( )μ ; μ ; μR G B are calculated that correspond to the
colour channels , , R G B.The value for each channel
is computed using the weight coefficients for every
block. For the k-th block an average deviation of
the green channel can be evaluated using the for-
mula:
1
1
σ ( ) μ ( ).
σ ( )
p
G
Gp
k Gi
kStd Avg G k
i=
=
= ⋅∑
∑
(6)
Based on the obtained values the correction fac-
tors for each colour channel are evaluated:
,
3
Std Avg R Std Avg G Stv Avg Bcorr G
Std Avg G
+ +
=
⋅
,
3
Std Avg R Std Avg G Stv Avg Bcorr R
Std Avg R
+ +
=
⋅
(7)
.
3
Std Avg R Std Avg G Stv Avg Bcorr B
Std Avg B
+ +
=
⋅
4. White Patches in YCbCr Colour Space
This method was introduced by Wang, Chen and
Fuh [6]. The image is represented in the YCbCr
colour space. White balance correction factors can
be obtained from the following expressions:
max max max, , ,
avg avg avg
Y Y Ycorr R corr G corr B
R G B
= = =
(8)
where maxY — maximum intensity value in a given
colour space [7].
5. White Balance Automated Correction Process
In the process of shooting, the device (a camera) is
able to change the way of white balance correction
(this effect is achieved through a number of modes
built in to the algorithms of the camera).
The main advantage of this approach is its sim-
plicity of implementation, as well as optimality in
terms of the device usage. However, when shooting
in RAW format, or solving computer vision issues
aimed at searching the changes in the state of the
image, and performing several images or several
points of a specific image comparison in standard
custom formats, such white balance correction
implementation does not work properly (the white
balance correction factors do not tend to unity).
As a part of the image points representation
comparison, such white balance adjustments can
affect the loss of a certain number of colours, lea-
ding to failures in the accuracy of colour compo-
nents calculations which leads to an incorrect com-
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance
ISSN 2706-8145, Control Systems and Computers, 2019, № 4 31
parison result. As a result, the computation of the
distance between the representations of points on
the colour surface that is performed by the means
of the normalization process will return incorrect
values, since the camera’s white balance correc-
tion for the selected scene (selected image) was not
taken into account.
Proposed Approach
Let us review the process of assessing the differen-
ce between two points of the image working in
terms of the issue described in paragraph 2 of this
paper. The camera automatically adjusts the white
balance in accordance with one of the modes set in
the camera by default. The assessment of the diffe-
rence between two points of the image is performed
by normalization. The Euclidean norm is used as
the calculation method.
The incoming data for the calculation of the
norm is represented by the colour values of the
points obtained from the camera. At the same time,
the obtained values are converted from the colour
space of the camera to the traditional colour space
by considering the white balance correction fac-
tors that were used during the image capture. This
conversion allows to transform a polygon that is the
colour representation of the camera on the surface
into a polygon that is representation of the classical
RGB model.
This conversion uses the matrix method and has
the following structure:
1
11 12 13 1 2 1
21 22 23 2 3 2
31 32 33 3 3
,
m
m
m
m
x
C C C C x y
C C C C x y
C C C C y
x
⎛ ⎞
⎜ ⎟⎛ ⎞ ⎛ ⎞⎜ ⎟⎜ ⎟ ⎜ ⎟⎜ ⎟⋅ =⎜ ⎟ ⎜ ⎟⎜ ⎟⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠⎜ ⎟
⎜ ⎟
⎝ ⎠
�
�
� �
(9)
where matrix С is the matrix of conversion, that
considers the white balance correction in terms
of transition to the colour model RGB traditional
look, vector x — point representation in the camera
color space, vector у — the resulting representation
of a point in the RGB color space.
The calculations are performed in terms of
working with integer values, and also can be used
for direct implementation into the hardware logic
of devices. This fact makes it possible to reduce
both the overall cost of resources for the perfor-
ming operations, as compared to the existing float
arithmetic, and the requirements for the sensor
configuration, thereby increasing the overall level
of performance.
Thus, we obtain a complex normalization algo-
rithm that takes into account white balance correc-
tion factors, and can be used both in the devices
that perform the shooting and in computers for the
organization of computer vision systems.
Conclusion
In the absence of the possibility to calibrate sen-
sors directly during the use of the device, existing
methods of colour identification are not as efficient
as possible in terms of working with the computer
vision issues. This is due to the fact that each of
the devices (or rather, each of the sensors of this
device) "receives" the corresponding section of the
colour model, which is deformed relative to the
similar standardized section allocated to the RGB
model. In addition, in the process of shooting the
device performs white balance adjustment based
on the parameters set for it by default (as a set of
automatic correction modes). On the basis of such
parameters as the scene illumination, the position
of the shot object, the position of the device rela-
tive to the scene, – for each of the frames its own
correction is formed which entails another defor-
mation of the section of the colour surface corre-
sponding to the sensors of the device. As a result,
the accuracy of image point color representation
assessment analysis drops significantly, making it
impossible to apply standard methods of compu-
ting the differences between two pixels by perfor-
ming normalization since the actual positions of
their representations on the colour surface differ
from the positions of the representations identi-
fied by the device. These facts are also supported
by the situation where the implementation of the
float arithmetic in the algorithms for calculating
the norm in the hardware logic of sensors, is unac-
ceptable since it significantly increases the overall
costs not only for the norm computation, but also
increases the level of requirements for the sensor
M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko
32 ISSN 2706-8145, Системи керування та комп'ютери, 2019, № 4
REFERENCES
1. Sinha, R.K., Pandey, R., Pattnaik, R., 2017. “Deep Learning for Computer Vision Tasks: A review”. Int. Conf. on Intel-
ligent Computing and Control (I2C2), pp. 1–5, https://arxiv.org/ftp/arxiv/papers/1804/1804.03928.pdf.
2. Stokes, M., Anderson, M., Chandrasekar, S., Motta, R., 1996. A Standard Default Color Space for the Internet — sRGB.
[online] Available at: <https://www.w3.org/Graphics/Color/sRGB.html> [Accessed 15 Apr. 2019].
3. Precise measurements are vital to colour sensor sensitivity”. EET India. [online] Available at: <https://www.embedded.
com/design/mcus-processors-and-socs/4007122/Precise-measurements-are-the-key-to-color-sensor-sensitivity>
[Accessed 2 Feb. 2019].
4. Land, E.H., McCann, J.J., 1971. “Lightness and Retinex Theory”. Journal of the Optical Society of America, 61(1),
pp. 1–11.
5. Lam, Hong-Kwai, Au, Oscar C., Wong, Chi-Wah, 2004. “Automatic white balancing using standard deviation of RGB
components”. ISCAS ’04, DOI: 10.1109/ISCAS.2004.1328898.
6. Weng, Ching-Chih, Chen, Homer, Fuh, Chiou-Shann, 2005. “A Novel Automatic White Patch Method for Digital Cam-
eras”. ISCAS 2005, DOI: 10.1109/ISCAS.2005.1465458.
7. Zapryanov, G., Nikolova, I., 2012. Automatic White Balance Algorithms for Digital Still Cameras – a Comparative
Study. Information Technologies and Control, 1, pp. 16-22.
Received 18.06.2019
equipment, causing a significant drop in perfor-
mance of normalizing algorithms due to the im-
possibility of their parallelization.
The approach proposed in the article, aims to
eliminate these drawbacks. Colours comparison in
terms of using a standardized colour model (gamut)
allows not only to obtain the same results regardless
of the applied device, but also makes it possible to
form a comparative analysis of the devices. This ef-
fect is achieved by the means of the data conversion
that is obtained from the camera after performing
automatic white balance correction to the tradi-
tional colour model. This allows to increase the ac-
curacy of measurements in solving computer vision
issues aimed at comparing points of one image or at
a full comparison of two or a series of images.
The algorithm can be wielded both in devices
that shoot, and in computers that are used to per-
form image processing operations in computer vi-
sion systems.
Integer Norm for Difference Assessment of the Frame Elements Considering the White Balance
ISSN 2706-8145, Control Systems and Computers, 2019, № 4 33
М.В. Іващенко, студент, факультет прикладної математики, Національний техн. ун-т України
“Київський політехнічний інститут імені Ігоря Сікорського” (НТУУ «КПІ ім. І. Сікорського»),
просп. Перемоги, 37, Київ, 03056, Україна,
mivaschenko_51@lll.kpi.ua
Д.Д. Охримчук, студент, факультет прикладної математики, НТУУ «КПІ ім. І. Сікорського»,
просп. Перемоги, 37, Київ, 03056, Україна,
den5096@gmail.com
Л.А. Люшенко, кандидат технічних наук, старший викладач,
факультет прикладної математики,
НТУУ «КПІ ім. І. Сікорського», просп. Перемоги, 37, Київ, 03056, Україна,
lyushenkol@gmail. com
КОНЦЕПЦІЯ ЦІЛОЧИСЛОВОЇ НОРМИ ОЦІНКИ РІЗНИЦІ
МІЖ ЕЛЕМЕНТАМИ ЗОБРАЖЕННЯ З ВРАХУВАННЯМ БАЛАНСУ БІЛОГО
Вступ. Оцінка різниці між двома елементами зображення є важливим блоком всіх алгоритмів графічної обробки.
Проте на сьогоднішній день більшість норм, які визначаються за колірними просторами, виконують операції з
плаваючою точкою і без врахування внутрішньої структури колірної моделі камери.
Мета. Метою даної статті є дослідження запропонованого підходу, на основі якого можливо виконати синтез
цілочислової норми, що враховує баланс білого камери при розрахуванні оцінки різниці між елементами зобра-
ження.
Методи. Камера автоматично налаштовує баланс білого відповідно до одного з режимів, встановлених за за-
мовчуванням. Оцінка різниці між двома точками зображення здійснюється за рахунок виконання нормалізації.
Як метод розрахунку використовується евклідова норма.
Результати. Запропоновано підхід до нормалізації, що враховує коефіцієнти корекції балансу білого. Розрахун-
ки виконуються з точки зору здійснення операцій з цілими значеннями, що надає можливість їх використання
для безпосереднього розгортання всередині апаратної логіки пристроїв. Цей факт дозволяє знизити як загальну
вартість ресурсів для виконуваних операцій, по відношенню до існуючої арифметики з плаваючою крапкою, так і
зниження вимог до конфігурації датчиків.
Запропонований підхід дозволяє підвищити ефективність вимірювань при вирішенні задач комп'ютерного
зору, спрямованих на порівняння точок одного зображення або при повному порівнянні відповідних елементів
двох або серії зображень.
Висновки. Запропонований у статті підхід спрямований на усунення недоліків сучасних систем, що викорис-
товують оцінку кольорів. Алгоритм може використовуватися як в пристроях, які здійснюють відеозйомку, так і в
комп'ютерах, які використовуються для виконання операцій обробки зображень в системах комп'ютерного зору.
Ключові слова: комп’ютерний зір, корекція балансу білого, датчики кольору, колірна модель, опорні кольори, норма
точки зображення.
M.V. Ivashchenko, D.D. Okhrymchuk, L.A. Lyushenko
34 ISSN 2706-8145, Системи керування та комп'ютери, 2019, № 4
М.В. Иващенко, студент, факультет прикладной математики, Национальный техн. ун-т Украины
«Киевский политехнический институт имени Игоря Сикорского» (НТУУ «КПИ им. И. Сикорского»), просп.
Победы, 37, Киев, 03056, Украина,
mivaschenko_51@lll.kpi.ua
Д.Д. Охримчук, студент, факультет прикладной математики, НТУУ «КПИ им. И. Сикорского», просп. Победы, 37,
Киев, 03056, Украина,
den5096@gmail.com
Л.А. Люшенко, кандидат технических наук, старший преподаватель, кафедра программного обеспечения
компьютерных систем, факультет прикладной математики (ПОКС ФПМ),
НТУУ «КПИ им. И. Сикорского», просп. Победы, 37, Киев, 03056, Украина,
lyushenkol@gmail. com
КОНЦЕПЦИЯ ЦЕЛОЧИСЛЕННОЙ НОРМЫ ОЦЕНКИ РАЗНИЦЫ
МЕЖДУ ЭЛЕМЕНТАМИ ИЗОБРАЖЕНИЯ С УЧЕТОМ БАЛАНСА БЕЛОГО
Введение. Оценка разницы между двумя элементами изображения является важным блоком всех алгоритмов
графической обработки. Однако на сегодняшний день большинство норм, которые определяются по цветовым
пространствам, выполняют операции с плавающей точкой и без учета внутренней структуры цветовой модели
камеры.
Цель. Целью данной статьи является исследование подхода к синтезу целочисленной нормы, которая учитывает
баланс белого камеры при расчете оценки разницы между элементами изображения.
Методы. Камера автоматически настраивает баланс белого в соответствии с одним из режимов, установленных
по умолчанию. Оценка разницы между двумя точками изображения осуществляется путем выполнения нормали-
зации. В качестве метода расчета используется евклидова норма.
Результаты. Предложен подход нормализации, учитывающий коэффициенты коррекции баланса белого.
Расчеты выполняются исходя из операций с целыми значениями, что дает возможность использования их для
непосредственного развертывания внутри аппаратной логики устройств. Это позволяет снизить как общую сто-
имость ресурсов для выполняемых операций, по отношению к существующей арифметике с плавающей точкой,
так и упростить требования к конфигурации датчиков.
Это повышает эффективность измерений при решении задач компьютерного зрения, направленных на срав-
нение точек одного изображения или при полном сравнении соответствующих элементов двух или серии изобра-
жений.
Выводы. Предложенный в статье подход направлен на устранение недостатков современных систем, исполь-
зующих оценку цветов. Алгоритм может использоваться как в устройствах, видеосъемки, так и в компьютерах,
используемых для выполнения операций обработки изображений в системах компьютерного зрения.
Ключевые слова: компьютерное зрение, коррекция баланса белого, датчики цвета, цветовая модель, опорные цвета,
норма точки изображения.
|