Для контакту з
авторами дивіться їх список.
To contact the authors look their list.
LI. Проблеми створення образного комп’ютера
Problems of
Pattern Computer Designing
1.
Гриценко В.І., Вінцюк Т.К., Федорин
Я.В. (Київ,
Україна)
² ДНТП “Образний комп’ютер”: здобутки, проблеми, перспективи
Hrytsenko V.I.,
Vintsiuk T.K., Fedoryn Ja.V. (Kyjiv, Ukraine) ² State R&D
Program “Pattern Computer”: Advances, Problems, Prospectives.
2.
Тарас Вінцюк (Київ, Україна) ² Генеративна
модель образного комп’ютера
Taras K. Vintsiuk (Kyjiv, Ukraine) ² Generative Model for Pattern Computer.
Pattern computer (PatCom) is such intellectual cybernetic system that provides a functional simulation of intelligent, mainly subconscious activity of all living and a human being particularly. In PatCom this activity is related to image, sound and other patterns perception, scene analysis, action and movement planning, generalisation of observations, discovering of regularities, prediction, decision making etc. Pattern computer operates with patterns and other complex notions. It actualises both pattern and logical reasoning.
Pattern computer is a parallel system. It has several information perception channels (acoustic, visual, and scential) that is multimodal perception, pattern operation system, improved human-machine interface. Unlike usual computer, which is based on a rapid arithmetic-logical processor and a large RAM space, PatCom is grounded on external world models including physical, geometry, acoustical, language, linguistic, semantical, canonical forms etc models.
Here is a so-called generative model for how to create a pattern computer is debated. This model is based on both the generative grammar hierarchy for multimodal prototype scene composition and the comparison of them with a scene to be perceived. So a multimodal informаtion synthesis is used as a feedback in the pattern analysis and understanding. Examples are given in reference to dictation and spoken translation machine creating and scene analysis.
3.
Schlesinger M. I. (Kyjiv, Ukraine), Flach B. (Drezden, Germany) ² Analysis of Optimal Labelling Problems and Their Application to Image Segmentation and Binocular Stereovision
Шлезінгер
М. І. (Київ, Україна), Флах Б. (Дрезден, Німеччина) ² Аналіз проблем оптимальної розмітки та їх
застосування в сегментації зображень та бінокулярному стереобаченні.
4.
Ярема Зєлик, Михайло
Личак, Олександр Лук’янчук (Київ,
Україна) ² Сприйняття і аналіз акустичних сигналів,
розпізнавання типу джерела звуку, оцінювання напрямку та відстані до джерела
Yarema Zyelyk, Mykhajlo Lychak, Oleksandr Lukianchuk (Kyjiv, Ukraine) ² Perception and Analysis of Acoustic
Signals, Recognition of Sound Source Type and Estimation of Direction and
Distance to a Source.
The initial results of researches obtained within the framework of State scientific-technological program "Pattern Computer" are considered. The realistic mathematical models of an acoustic component of the external world are offered. The typical sources of an acoustic noise in aerospace engineering are analyzed. The structure of a receiving acoustic antenna and choice of input-output device of acoustic signals in the computer is justified. The functional purpose of the designed software in MATLAB is described. The results of inputting, analysis and representation in the forms of vector time series, matrices of spectral densities and matrices of spectrograms (matrices of patterns!) of real acoustic sources signals in the room are represented. The technique of estimation of a direction and distance to a sound source is offered and the examples of its application are considered.
5.
Юрій Крак, Тарас
Вінцюк, Микола Кириченко, Федір Гаращенко, Олександр Бармак (Київ, Україна) ²Розробка комп’ютерних технологій
моделювання та керування візуальними образами людського обличчя при синтезі
мовлення
Yuriy Krak, Taras Vintsiuk, Mykola Kyrychenko, Fedir Garashchenko, Olexander Barmak (Kyjiv, Ukraine) ² Development of Computer Technologies for
Modeling and Control of Visual Images of Human Face Under Speech Synthesis.
Technologies for modeling of processes animation of arbitrary text
of are
proposed. Mathematical model of
human head with possibilities of facial expression during conversation and
synchronization with speech are created.
6.
Олег Березький (Тернопіль, Україна) ² Підходи до опису
і синтезу симетричних зображень
Oleh Berez’kyj (Ternopil’,
Ukraine) ² Approaches to Symmetrical Image
Description and Synthesis.
One of ways to construct images recognition algorithms is a
generating model which eves in reproducing of images of a given class and
further automatic comparing of the studied image with the generated one. A
class of complex symmetric images (images-ornaments) is considered in the
report. Classes of images bearing symmetric structure are widely represented in
nature, science, art. The constituents as ornament, sub-ornament, rapport and
elementary image. To describe the latter a language of description consisting
of a number of non-derivative elements and series of single and binary
operations between them. The report represents two approaches to generating
symmetric images: matrix and recursive. Using the recursive scheme to groups of
transformations there are recursive models for symmetric groups in space and
stripe obtained. Matrix models of synthesis are based on transformations matrix
performing reflection, rotation and displacing. The suggested language and
algorithms of synthesis are used to construct ornaments editor.
¨ II. Теоретичні питання оброблення та розпізнавання
сигналів і зображень
Theoretical Problems of Signal
Processing and Recognition
7.
Михайло Личак (Київ, Україна) ² Хаотичні події, процеси та їх характеристики
Mykhajlo Lychak (Kyjiv, Ukraine) ² Chaotic Events, Processes and Their Characteristics.
8.
В. І. Васильєв (Київ, Україна) ² Самоорганізація
колективів вирішувальних правил в задачах розпізнавання
V. I. Vasyl’jev (Kyjiv, Ukraine) ² Decision Rules Collectives Self-Organization in Recognition Tasks.
Розглянуто можливість
використання колективів вирішувальних правил в задачах розпізнавання образів.
Показано, що в алгоритмах, які базуються на методі граничних спрощень та методі
групового врахування аргументів, можуть самоорганізовуватись колективи
вирішувальних правил, до яких можуть бути застосовані всі відомі правила
формування колективних рішень.
9.
А. С.
Краснопоясовський, І. В. Шелехов (Суми, Україна) ² Оптимізація параметрів навчання за вибірками малого обсягу
A. S. Krasnopoyasovsky, I. V. Shelekhov (Sumy, Ukraine) ² Teaching Parameters Optimization based on Small Volume Samples.
The statement of the problem, mathematical model, criterion and algorithm of the teaching recognition space-temporary parameters systems optimization are considered within the framework of extreme-information method function-statistical tests (MFST). The characteristic feature of MFST is a possibility to correction during teaching an entrance mathematical description with a purpose of construction a classifier having a close to maximum asymptotic reliability of recognition.
10.
В. М. Корчинський (Дніпропетровськ, Україна) ² Інформативність афінно-інваріантної геометричної моделі проекційних
зображень в їх морфологічному аналізі
V. M. Korchyns’kyj (Dnipropetrovs’k, Ukraine) ² Informativeness of Affine-Invariant
Geometrical Model for the Projective Images in Their Morphological Analysis.
The characteristics (normalized cumulants of brightness
functions) and their linear combinations are determined, most informative for
position-invariant recognition of the projective images received in conditions
of distant photogrammetry. The technique is offered for the definition of
optimum dimension of model space for identification and morphological analysis
of such images with given by reliability. The
distributions of brightness of such images in a picture plane are represented
by points of varieties in model space, invariant as to item parameters of
projection (mutual orientation of subject and picture planes, hardware function
of the video gauge), and to a spectral range, in which the images are fixed.
The morphological analysis is reduced to the decision of item and metric tasks
of geometry of space of model.
11.
О. М. Ахметшин, Е.
В. Пирогов (Дніпропетровськ,
Україна) ² Кластеризація багатовимірних та
багатопараметрових зображень “без вчителя” на основі генетичного алгоритму
O. M. Akhmetshin, E. V. Pirogov (Dnipropetrovs’k, Ukraine) ² Unsupervised Clasterisation of Multidimensional
and Multiparameter Images on Base of Genetic Algorithm.
It is offered a new unsupervised clustering method of multidimensional and multiparameter data based on the genetic algorithm. The model is capable to an estimation of amount clusters in data. A basic self-organizing model is the genetic algorithm with a specialized operator of crossover, capable to improve results obtained on the last step of the iterative process. The application of such specialized operator has enabled constructions of a new method of clusterization. The paper consists of the theoretical description of the offered model and its parameters, examples of clustering multiparameter NMR medical image and multidimensional image of the earth surface.
12.
В. М. Хавалко (Львів, Україна) ² Синтез зображень,
що характеризуються точковими типами симетрії
V. M. Khavalko (L’viv, Ukraine) ² Image Synthesis, Characterized by Point Types of Symmetry.
The problem of synthesis of image and complex geometrical objects (CGO) which characterized by point types of symmetry is reviewed in this report. The synthesizing process is based on usage of the functional and recursive approaches. For extension of set of the synthesized images the using of fractal curves of different dimensions and levels of detailing are offered as the elementary objects. The main techniques of synthesizing of the symmetrical image on the base of R-functions are reviewed. On the base of conducted analysis of these techniques the algorithms of synthesis of CGO with point types of symmetry are offered. Information and analytical system of image synthesis include symmetrical image is developed and realized. The results of simulation are represented and the analysis of advantages and disadvantages of applying of the recursive and functional approaches to CGO synthesis is carried out.
13. Г. М. Востров, М. В. Полякова, В. В. Любченко (Одеса, Україна) ² Сегментація даних
з використанням вейвлет-аналізу
H. M. Vostrov, M. V. Poliakova, V. V. Liubchenko (Odesa, Ukraine) ² Data Segmentation Using Wavelet Analysis.
The paper is devoted to financial time series segmentation based on wavelet analysis A method of revealing of properties change of nonstationary time series is proposed. This method is based on a tie of local function regularity with wavelet coefficient values. However, unlike of the papers, where in each point a regularity index is calculated, we offer less labor-intensive approach, based on calculation of statistics from wavelet coefficients and them comparison with thresholds got by us. This segmentation method is used for preprocessing stage of cyclic analysis of financial time series.
14. Довженко Оксана Володимирівна (Дніпропетровськ, Україна) ² Сегментація текстурних зображень на основі методу векторної підтримки
Dovzhenko Oksana (Dnipropetrovs’k, Ukraine) ² Texture Images Segmentation Based on
Vector Support Method.
The aim of this work is to propose a texture segmentation methodology on the base of Support Vector Machines. This technology has two parts: 1. The main module is a SVM whose goal is the supervised image segmentation in order to detect and localize object having regular patterns. 2. A simple median-filtration and morphological postprocessing for detection and removing of the misclassificated pixel blocs and representation of the results for useful visual evaluation. The obtained result show that SVM approach provides the good results for texture segmentation problem for different types of texture images.
15.
Р. П. Базилевич, І.
В. Подольский, П. Р. Базилевич (Львів, Україна) ² Декомпозиція
складних схем на основі ієрахічної кластеризації
R. P. Bazylevych, I. V. Podol’s’kyj, P. R. Bazylevych (L’viv, Ukraine) ² Decomposition of Complex Circuit by
Hierarchical Clustering.
The algorithms for high and very high size circuit’s decomposition are proposed. The Optimal Circuit Reduction method is used as a methodological basis. The method allows recognizing mutually inserted circuit’s clusters. Some algorithms for initial solution and its optimization are suggested. For initial solution as a first stage we use the Optimal Reduction Tree. Three strategies are possible next: serial, parallel-serial and dichotomous. We develop and recommend using the latter for the better quality of initial solution. During the optimization the clusters of arbitrary size are used for transferring and exchanging between partitions instead of original circuit elements. It drastically reduces the computation complexity and improves the quality of solution. To escape from the local extreme some strategies are investigated. The experimental results confirm the effectiveness of developed methodology on the well-known IBM test-cases.
16.
Надія Тимофієва (Київ, Україна) ² Дві задачі розбиття
Nadija Tymofijeva. ² Two Partition Problems.
O ІІІ.
Усномовні інформаційні технології
Automatic Recognition and Synthesis of Speech Signals
17.
Konstantin Biatov, Martha Larson, Stefan Eickeler (Sankt Augustin, Germany) ² Zero-Crossing-Based Temporal Segmentation and Classification of Audio Signals
Костянтин
Біатов, Марта Ларсон, Стефан Ейкелер (Санкт Августин, Німеччина) ² Темпоральна сегментація та класифікація
аудіосигналів на основі нуль-переходів.
The task of audio classification is streamlined if the audio stream can be decomposed in real time into homogeneous segments larger than the individual frame. If segment morphology reflects the underlying audio type, classifier design and training becomes more transparent and can directly exploit real-world knowledge. The experiments presented here explore real-time zero-crossing-based audio stream segmentation and classification. We segment an unedited audio stream from a German-language radio program into three classes: speech, speaker breathing and music. Features related to modulation of the amplitude envelopes are exploited to segment the audio stream into homogeneous sub-syllable segments. The segments prove useful for classifying the audio-stream, especially the speech portions. We use features extracted from zero-crossing-based pseudo-spectrum to train a Gaussian classifier that classifies audio content into the three categories. Our results demonstrate that our zero-crossing-based segmentation and classification method is viable one, yielding satisfactory performance in real time.
18.
В. Г. Грибенко (Яготин, Україна) ² Детектор звукових образів, що емітуються слуховим органом людини
V. H. Hrybenko (Jahotyn, Ukraine) ² Detector of Audio Patterns Emitted by
Human Ear Organ.
19.
Тарас Вінцюк, Микола
Сажок, Тетяна Людовик, Руслан Селюх (Київ, Україна) ² Автоматичний озвучувач українських текстів на основі фонемно-трифонної
моделі з використанням природного мовного сигналу
Taras Vintsiuk, Mykola Sazhok, Tetyana Lyudovyk, Ruslan Selyukh (Kyjiv, Ukraine) ² Automatic Ukrainian Text-to-Speech System
Based on Phoneme-Threephone Model Using Natural Spoken Signal.
The text-to-speech system in time domain for Ukrainian is described. The concatenated acoustic elements are chosen in accordance to phoneme-threephone model for speech synthesis. Acoustical data is taken from the speaker voice passport. The computerized tools for speech synthesis research and development are described.
20.
Daniel Tihelka, Jindřich Matoušek, Martin Vlach (Plzeń, Czech Republic) ² Modelling Word Stress for Use in Speech Synthesis
Даніел
Тігелка, Їндржих Матушек, Мартін Влах (Плзень,
Чехія) ² Моделювання наголосів у словах для
використання при синтезі мовлення.
This paper deals with word stress modeling for use in text-to-speech systems. It describes one of the ways of stress modeling. Several experiments were carried out for different phonemes containing stress and other prosodic features. Listening tests with specially designed words were set up for the comparison of speech generated from each experiment. The results show that the examined way of stress modeling can increase intelligibility and naturalness of the synthetic speech.
21.
Pavel Brodsky, Luděk Müller (Plzeń, Czech Republic) ² Automatic Phonetic Baseforms for Person Names in the Czech Dialogue System
Павел
Бродски, Людєк Мюллер (Плзень, Чехія) ² Автоматичні фонетичні базові форми для
імен осіб у чеськомовній діалоговій системі.
Phonetic baseforms are the basic recognition units in most speech recognition systems. Linguists usually determine these baseforms once a vocabulary is chosen and not modified thereafter. However, several applications, such as name dialing require the user to be able to add new words to the vocabulary. These new words are often names or task-specific jargons that have user dependent pronunciation. In this paper, we describe a method how to generate a phonetic baseform from acoustic pronunciation of a name without a prior knowledge of the name spelling. We used a language model based on bigram statistics.
22.
Jan Romportl, Jindřich Matoušek, Daniel Tihelka (Plzeń, Czech Republic) ² Prosody Model and Its Application to Czech TTS System
Ян
Ромпортл, Їндржих Матушек, Даніел Тігелка (Плзень,
Чехія) ² Модель просодики та її застосування у
чеськомовній системі озвучення текстів.
The first part of this paper
proposes a formal theoretical framework for prosody description. This framework
is based on empirically acquired axioms (following the linguistic
structuralism) and in terms of the mathematical set theory it presents prosody
as a relation between abstract sentence underlying structures and intonation
(together with timing). On the basis of this framework one can utilize various
system description and analysis methods as well as pattern processing
techniques to model the aforementioned relation. The second part of this text
introduces the application of this framework to the Czech text-to-speech system
ARTIC. It uses rules to
place abstract intonation schemes (melodemes and cadences) depending on the
position of an intonation centre of an utterance.
23.
Зореслава Шпак, Юрій
Рашкевич (Львів, Україна) ² Аналіз темпоральних перетворень мовних елементів для задач часового
масштабування голосових повідомлень
Zoreslava Shpak, Jurij Rashkevych (L’viv, Ukraine) ² Spoken Elements Temporal Transformation
Analysis for Temporal Spoken Messages Scaling Tasks.
In order to provide the high intelligibility and naturalness of the voice rate-change records the natural regularities of the oral speech temporal processes must be considered in the time-scale modification methods. The characteristics of the speech units (syntagmas, words, syllables, etc.) duration have been specified for different pronunciation types. Time-scale transformation regularity of the Ukrainian voice phonemes sounding at different speech rates has been investigated. Six groups of oral speech elements are determined: stressed vowels, unstressed vowels, voiced consonants, unvoiced consonants, plosive stops and word-spacing pauses. The elements of specified groups are characterized by similar temporal transformations. Non-linear analytic functions presenting the relative temporal changes of sound duration have been plotted for every group. It was established that the main spectral characteristics of voice signals, particularly, pitch and formant frequencies are practically independent of speech rates.
ï@ IV. Технічна діагностика об’єктів і машин за
їх сигналами та полями
Automatic Diagnostic of Objects and
Machines on Its Signals and Fields
24.
Юзефович Р. М.,
Михайлишин В. Ю. (Львів,
Україна), Яворський І. М. (Львів, Україна; Бидгощ, Польща) ² Когерентний кореляційний аналіз стохастичних циклічних навантажень
Youzefovych R. M., Mykhailyshyn V. Ju. (L’viv, Ukraine), Javors’kyj I. M. (L’viv, Ukraine;
Bydgość, Poland) ² Coherent Correlation Analysis of
Stochastic Cyclic Loadings.
Properties of probabilistic
characteristic coherent estimations of periodically correlated random processes
(PCRP) are considered. The probabilistic structures of cyclic loadings of
bearing knot vibrations with ununiform radial tension are investigated.
25.
Ігор Ісаєв, Георгій
Трохим, Ігор Яворський (Львів,
Україна) ² Властивості компонентів дисперсії
імовірнісної моделі сигналу електромагнітного збурення рейок залізничної колії
Ihor Isayev, Georgij Trokhym, Ihor Javors’kyj (L’viv, Ukraine) ² Variance Components Properties of
Probabilistic Signal Model of Rail Electromagnetic Excitation.
In this paper power properties of rail flaw detection signal based of probabilistic model in assumption of periodical correlated stochastic process (PCSP) are investigated. Expressions for variance components of Fourier series expansion by methods of nonstationary statistical processing were analyzed. Variance components content according to "defect less" rail model was explored. Similar components of real rail signals without defects and with two defects are presented too.
26.
Р. Я. Косаревич, Б.
П. Русин (Львів, Україна) ² Сегментація напівтонових зображень мікроструктури матеріалів для визначення
динаміки росту кавітаційних тріщин
R. Ja. Kosarevych, B. P. Rusyn (L’viv, Ukraine) ² Segmentation of Materials Microstructure
Greyscale Images for Cavity Cracks’ Growth Rating.
The problem of image segmentation of surface damaged by cavity
cracks is considered. The objective of such investigations is calculation the
cracks quantity and dimension. At first stage the points, which satisfied the
proposed criterion, is selected as seeded regions. Next adjacent points are
joining under offered seeded region-growing technique. At last stage among
segmented objects rejected those, which correspond to microsections.
27. В. Д.
Павленко, О. О. Фомін (Одеса, Україна) ² Стиснення даних при
формуванні простору діагностичних ознак на основі нелінійних непараметричних моделей
об’єктів контролю
V. D. Pavlenko,
O. O. Fomin (Odesa, Ukraine) ² Data Compression in Forming of Diagnostic
Features Space on Base of Nonlinear Nonparametrical Control Objects Models.
As informing control objects description offers to use nonlinear nonparametric dynamic models as Volterra rows. Herewith aggregate of polymeasure gravimetric functions - Volterra kernels (VK) - fully describes nonlinear and dynamic properties, and consequently, control objects technical fortune. Diagnostic procedure in this case boils down to decision of control objects nonparametric authentication task - to determination VK on data of experiment “entrance - exit", and to construction on base of got kernels of diagnostic signs system, in space of which a decisive regulations of optimum classification is built.
It Got signs informating estimations, formed on base of main section VK of second order for considered control object: heuristic signs, moments VK , and also signs, picked out by compression method, founded on Wavelet-transformation.
28. В. Ф. Миргород, І. М. Гвоздєва, С. М. Ковешніков (Одеса, Україна) ² Формування функції просторового розрізнення розподіленої мережі датчиків у системі неруйнуючого контролю
V. F. Myrhorod,
I. M. Hvozdeva, S. M. Koveshnikov (Odesa, Ukraine) ² Distributed Sensor Array Space-Resolution
Function Forming in the Nondestructive Diagnostics.
Structured and parametric synthesis method based on the approximation of space-resolution function by truncated Fourier-Bessel series for the distributed axle-symmetrical signal-processor sensor array in the ultrasonic nondestructive diagnostics are proposed.
29. Андрій Павлович Гвоздак, Оксана Володимирівна Довженко (Дніпропетровськ, Україна) ² Виявлення порушень структури подовжених об’єктів на основі час-частотних перетворень з перерозподіленням
Andrij P.
Hvozdak, Oksana V. Dovzhenko (Dnipropetrovs’k, Ukraine) ² Extended Object Structure Destruction
Detection Based on Time-Frequency Transformations with Reassignment.
One of the power instruments for analysis non-stationary signal is time-frequency distribution Cohan's class with reassignment. Using the reassignment principle provides efficiently combine a reduction of the interference terms and well adapted smoothing kernel and an increased concentration of the signal components achieved by the reassignment. The principle is to raise much reliability of analysis time-frequency mapping images. Also present using time-frequency distribution with reassignment for detection structural changes at non-destructive testing hot rolling pipes by use magnetic-inductive transducer system. The distributions such type allows raise more reliability of testing and very pliable adjust the threshold of sensitivity by change length smoothing windows in time and frequency direction instead of set magnitude parameters. In implementation part it is introduced reassignment time-frequency distributions applied to signals recorded from magnetic transducer system in non-destructive testing of carbon boiler pipes. Analysis time-frequency plane of distributions with different base kernel can say that this method provides higher concentration in the time-frequency plane, but does not remove all cross terms.
N V.
Автоматичне оброблення та розпізнавання зображень
Automatic Image Processing and Recognition
30.
Муригін К. В. (Донецьк, Україна) ² Виявлення ключових точок на зображенні обличчя людини
Muryhin K. V. (Donets’k, Ukraine) ² Detection of Essential Points on Human
Face Images.
In this work, the localization algorithms of the essential facial points are described. Presented methods base on the correlation matching of the local feature vectors based on a wavelet transform, which is defined as a convolution with a family of complex Gabor kernels. The essential facial points were located on the face image in the form of rectangular grid (face image graph). The presented algorithms were tested on the face image database that consists of 200 frontal images of different persons. The merits and demerits of the described methods and results of theirs comparison are produced.
31.
Копитчук М. Б.,
Олещук О. Б. (Одеса,
Україна) ² Попередня обробка зображень
Kopytchuk M. B., Oleshchuk O. B. (Odesa, Ukraine) ² Image Pre-Processing.
In practice the task of predefined kind objects selection from photos often appears. As an example of such objects the textual inscriptions are considered. As a criterion for typical features formation the geometrical sizes (external dimensions and line thickness) are used. The principles of image division on symbols and irrelevant objects are based on the supposition that in the small regions, which are directly side with symbols, brightness and contrast are close to maximum. In the paper it is shown that one of the first steps of image analysis must be background extraction, brightness, and contrast equalization along the whole of image. To accomplish above actions a number of matrix operations is introduced, namely maximum distribution, minimum distribution and region normalization. The optimized algorithm of introduced operations realization is proposed. The algorithm allows greatly reducing the computation cost. The result of image preprocessing is suitable for scene analysis and character recognition.
32.
Олександр Ільченко (Львів, Україна) ² Гістограмний метод розпізнавання рукописного тексту і символів
Oleksandr Il’chenko (L’viv, Ukraine) ² Histogram Method for Handwritten Text and Symbol Recognition.
In process described designed bar graph method philosophy of a building of a system of recognition handwritten to the text and numerals also generated. Bar graph method of handwriting recognition and numerals is based on construction of the schedule (histogram) of dependence of quantity {amount} of informative pixels of a template of the binary half-tone image postponed on an ordinate axis, from coordinate of a line or a column, postponed on an axis of abscissas and its comparison with a panel of reference histograms of initial sampling.
33. Hemappa B., Reddy N. V. S. (Davangere, India) ² Novel Segmentation and Feature Extraction Techniques for Handwritten Text Recognition
Гемаппа
Б., Редді Н. В. С. (Даванжер, Індія) ² Нові методики сегментації та виділення
ознак для розпізнавання рукописних текстів.
In this paper, we proposed novel slant detection and correction techniques, segmentation and feature extraction methods for handwritten text recognition. The segmentation and feature extraction module works on data, which is to be preprocessed. This is necessary because of the output of the scanned document is noisy and usually contains too many unwanted impression happened due to humidity and uneven ink distribution on paper. The novel techniques are used to detection and correction of any rotation that may occur during the writing or scanning process. The slant correction algorithm then generates a non_slant image by rotating the blocks, rather then the individual pixels. The proposed segmentation algorithm will count the block pixels in horizontal lines of the binarized image and estimate the width and base line of each text line in the document. This ensures easy to segment the document into text of sentences. Each sentence again segmented into words. Each word processed and extract the prominent feature for further classification. The proposed system based on neural network feature extraction gives promising results in terms of classification and recognition.
34.
В. В. Гнатушенко (Дніпропетровськ, Україна) ² Інваріантна модель оброблення та розпізнавання ізопланатичних зображень
V. V. Hnatushenko (Dnipropetrovs’k, Ukraine) ² Invariant Model for Processing and
Recognition of the Isoplanatic Images.
As it is known, the diffraction phenomena play the important role by
activity of optical instruments intended for obtaining of the objects’ images.
The images obtained with the help of focusing devices (of lenses), never
reconstruct object with ideal accuracy. They are screwed owing to any kind of
irregularities of optical systems (aberrations). But even the ideal lens, free
from aberrations, cannot give the ideal image through a wave nature of light.
Owing to diffraction phenomena the laws of ray optics are upset and the mapping
of points of object in points of the image ceases to be projective. In terms of
the theory of wave processes it means, that the image ceases to be stigmatic.
Thus, there is an important task of recovery and identification of such images,
which are not covered known informational geometric models.
35.
О. М. Ахметшин, М.
Г. Комарова (Дніпропетровськ,
Україна) ² Виділення інформативних ознак текстурних
зображень на основі метода рекурентного кількісного аналізу
O. M. Akhmetshin, M. H. Komarova (Dnipropetrovs’k, Ukraine) ² Application of the Recurrence
Qantification Analysis Method to Getting an Information Characteristics of
Texture Images.
The aim of this study is to investigate the
applicability of a recurrence quantification analysis (RQA) to the problem of
texture images and stochastic signals quantitative characteristics detection.
The proposed method is based on an idea of nonlinear dynamical data series
analysis in the embedding phase space. The field of nonlinear dynamics is
concerned with the way in which variables describing a system vary with respect
to each other instead of time. In many systems, however, it is only possible to
measure a single variable. The dynamics of such systems can be modelled by
applying the embedding theorem. The embedding theorem is based on the concept
that an observed scalar signal is a one-dimensional projection of the dynamics
of the system, which are represented in many dimensions. The embedding vectors are constructed from delayed samples of
the signal.
RQA is based on the computation
of an Euclidean distance matrix between the rows (epoch) of an embedded matrix
formed by the scalar time series at a fixed lag (time delay method). Then the
recurrence plot is built according to the rule: darken that elements of the
Euclidean distances matrix, which has a value less then some defined threshold.
These points are called 'recurrent points'. The darkened points individuate the
recurrences of the dynamical process and the plot can be considered as a global
picture of the autocorrelation structure of the system. Using taken plot the
five quantitative descriptors of RQA are calculated.
The technique was applied
to detection quantitative characteristics and to increase sensibility of the
texture images and x-ray images of the bone trabecula
structure, affected with osteoporosis.
The main advantage of proposed method is independence
of the technique on the statistics structure of data series, their length,
presence of noise and possibility to construct mathematical model of observed
data.
36. Ахметшин О. М., Киргизов І. О. (Дніпропетровськ, Україна) ² Сегментація низькоконтрастних зображень на основі методу мультифрактальної
декомпозиції
Akhmetshin O. M., Kyrhyzov I. O. (Dnipropetrovs’k, Ukraine) ² Low-Contrast Images Segmentation Based on
Multi-Fractal Decomposition Method.
The short theoretical lists are reduced which are connected to concept fractal-set with fractional Hausdorff dimension. The results and their analysis, and as application of the developed method multifractal of decomposition to the tasks of segmentation of textures and low-contrast images are shown. The algorithms for an evaluation of a multifractal decomposition method are represented.
37. Довженко Оксана Володимирівна, Ахметшин Олександр
Мубаркович (Дніпропетровськ,
Україна) ² Класифікація текстурних зображень в
інваріантному просторі моментів Цернакі аналітичних автокореляційних функцій
Dovzhenko Oksana, Akhmetshin Oleksandr (Dnipropetrovs’k, Ukraine) ² Texture Images Classification in Zernike Invariant
Moments Space of Analytic Autocorrelation Functions.
A new method of texture images classification is considered. The main idea of one is increasing sensitivity of new invariant features selection on base of analytical image conception. Into the bounds of new approach each analyzed texture is presented as complex image with analytical autocorrelation function. The last gives possibility for finding its magnitude and phase characteristics. In the next step obtained features are presented as rotation invariant pattern set of Zernike moments. To achieve scale and translation uniformity the regular moments of each feature image are utilized. Final classification of invariant characteristics is based on using of support vector machines. The obtained classification accuracy for 61-class data set is 99%. The considered images have difference in rotation. Thus one can conclude that the proposed features selection method are quite effective for image classification problem.
38.
Роман Воробель, Ігор Журавель, Костянтин
Сущик (Львів, Україна) ² Перетворення палітри кольорових зображень
Roman Vorobel’, Ihor Zhuravel, Kostiantyn Sushchyk (L’viv, Ukraine) ² Color Images Palette Transformations.
VІ. Цифрова фільтрація, кодування та відновлення
сигналів і зображень
Digital Filtering,
Analysis, Coding and Reconstruction of Signals and Images
39.
В. П. Савчук, І. М.
Глушко (Одеса, Україна) ² Керування п’єзоелектричною компенсацією вимірювальної камери за допомогою
нестаціонарного фільтра Калмана
V. P. Savchuk, I. M. Hlushko (Odesa, Ukraine) ² Control by Piezoelectric Compensation of
Measuring Chamber by the Medium of Non-Stationary Kalman Filter.
The problem of a non-stationary filtration of response of a composite vibrating system on change of signal's amplitude and phase of its excitation is considered. The solution of this problem is reduced to minimization of a functional loss, which depends on system state, using a Kalman filter. As a result the control quality of piezoelectric compensation of vibrations owing of decreasing effect of transient processes is improved.
40.
Sergey K. Abramov, Vladimir V. Lukin (Kharkiv, Ukraine) ² Adaptive Myriad Filters for 1-D Signal Processing
Сергій
Абрамов, Володимир Лукін (Харків, Україна) ² Адаптивні множинні фільтри для оброблення
одновимірних сигналів.
The generalized adaptive myriad filter framework is proposed and
the technique for such filter design is described. Within this framework and by
applying two new local activity indicators, the novel versions of locally
adaptive myriad filter are designed and tested. The appropriately good
performance of these new filters is demonstrated for the case of intensive
mixed additive and impulsive noise. The recommendations concerning the filter parameters
selection are presented.
41.
Мусатенко Ю. С.,
Курашов В. Н. (Київ,
Україна) ² Хвилькове та Карунена-Лоева перетворення
зображень у реальному часі
Musatenko Yu. S., Kurashov V. N. (Kyjiv, Ukraine) ² Wavelet and Karhunen-Loeve Image Transformation for Real-Time Applications.
В роботі
представлений розроблений авторами програмний пакет для здійснення хвилькового
(wavelet) та Карунена-Лоева(КЛ) перетворень зображень в реальному часі на
персональному комп’ютері. Для хвилькового перетворення використовався відомий
біортогональний фільтр 9/7, який є одним з найкращих з точки зору використання
в задачах стиснення з втратами.При розробці пакету максимальна увага
приділялась отриманню високої швидкодії. Для забезпечення високих швидкісних
показників реалізацію перетворень було здійснено в цілих числах, з
використанням технології ММХ та з спеціальною оптимізацією кешування даних.
Досягнутий виграш порівняно з традиційною реалізацією є 48 разів. Отримана
швидкодія тривимірного (КЛ х 2D Хвилькове) перетворення є 15 кадрів на секунду
для ансамблю 16 зображень 768х768 на ПК Pentium III 1GHz. Якість відновлення
зображень відповідає середньому піковому відношенню сигнал/шум ~50 дБ, що
достатньо для систем стиснення зображень з втратами, де вимоги до точності
відновлення, як правило, менше 40 дБ. Розроблений пакет призначений для
забезпечення роботи в реальному часі системи стиснення наборів зображень,
розробленої авторами раніше.
42.
Олег Капшій, Богдан
Русин (Львів, Україна) ² Система стиску дактилоскопічних зображень на основі вейвлетного
перетворення
Oleh Kapshij, Bohdan Rusyn (Kyjiv, Ukraine) ² Wavelet Based Fingerprint Compression
System.
The fingerprint compression task is considered. A loss compression scheme is chosen. This allows the high compression rate to be achieved. The structural scheme of the system includes: - a wavelet transformer that uses biorthogonal filters; - a uniform quantizer with dead zone; - a Lynch-Davisson entropy coder. A variable-length-to-block coding scheme is used that allows the entropy coder compression rate to be improved. The system software implementation was created and tested. The results comparison, when the entropy coder was changed to standard archiving utilities, is presented.
Compress rate improvement, based on information about distribution
of values of wavelet bands after quantizer and on fingerprint image modeling,
is proposed.
43.
О. В. Антонов, Ю. М.
Герасимов, М. В. Руженцев (Харків,
Україна) ² Підвищення якості відновлення зображення
імпульсних протяжних об’єктів
O. V. Antonov, Ju. M. Gerasimov, M. V. Ruzhentsev (Kharkiv, Ukraine) ² Improving of Quality for Image Reconstruction of Extended Impulse Objects.
The images of investigated objects are the aim of observation in many areas of science and engineering. At that to improve a quality of registered images its retrieving by non-linear methods is carried out taking into account different a priori information. The most simple among other equivalent methods is the iterative non-linear method with limitations. However the direct application of this method leads to some parasitic oscillations on surface of reconstructed image when impulsive-extended objects are retrieving. The causes of given effect origin is surveyed and the path its elimination is defined in this paper. For this purpose achieving the convolution of object main component is deducted from registered image and only after that the procedure of image restoring is carrying out.
44.
О. М. Ахметшин, О.
О. Киргизов (Дніпропетровськ,
Україна) ² Баєсовські нейронні мережі: порівняльний
аналіз точності класифікації
Akhmetshin O. M., Kyrhyzov O. O. (Dnipropetrovs’k, Ukraine) ² Bayesian Neuron Networks: Classification Accuracy Comparative Analysis.
This
paper introduces theoretical concepts Bayesian
end Probabilitic neural networks based on use Bayesian rule. Accuracy of classification and average probability of error are presented
for comare of networks – primarily Simple Bayesian networks and Bayesian multi-nets for classification tasks of medical data. The algorithms for an evaluation of a
Bayesian and Probabilitic NN’s are presented.
45. Т. Б. Мартинюк, В. В. Хом’юк, А. В. Кожем’яко, Н. В.
Фофанова, О. Б. Мартинюк (Вінниця,
Україна) ² Сортувальна нейроподібна мережа
T. B. Martyniuk, V. V. Khomyuk, A. V. Kozhemiako, N. V. Fofanova, O. B. Martyniuk. (Vinnytsia, Ukraine) ² Sorting
Neural-Like Network.
Recently the sphere of effective application of neural and neural-like networks quickly extends and covers such intensively developing areas, as signal/image processing and analysis, and images recognition. Simultaneously, the structure and organization of these networks are transformed and improved for relocation to new procedures and algorithms realization. All kinds and versions of neural and neural-like networks are detail described and classified in many references. Simultaneously, features of algorithm are correct known structures of networks. The example of algorithm implementation of numeric data sorting by the pair exchange method with a count on the neural-like networks is presented in the given work. The structure offered neural-like network represents as a version of bi-directional associative memory (BAM), which is the further development of Hopfield networks. The offered network with specific organization is named the S-network because it is oriented on performance of sorting procedure. Implementation of the S-network is supposed by optoelectronic element base.
46.
Volodymyr Borovytsky (Kyjiv, Ukraine)
² Fidelity and Correlation Measures of
Performance of Digital Imaging Systems
Володимир Боровицький (Київ, Україна) ² Вимірювання ефективності та якості
цифрових оптоелектронних систем.
47.
Роман Воробель, Олена Берегуляк (Львів, Україна) ² Двоточковий метод підвищення контрастності зображень
Roman Vorobel’,
Olena Berehuliak (L’viv, Ukraine) ² Two-Point Method
for Image Enhancement.
The analysis of contrast enhancement for gray-scale monochrome images with the usage of
local contrast enhancement under its linear description is realised. The
methods of this class use local contrast determination, its non-linear
enhancement, and reconstruction of improved image. Two-point method of contrast
enhancement of images based on definition of the sum of two neighbour elements
contrasts relative to local background is proposed and proved. Example which
illustrates the efficiency of proposed method by the way of generalised
contrast enhancement is submitted.
48.
О.М.Сухоручкіна (Київ, Україна),
В.Е.Прянічніков (Москва,
Росія), М.В.Прогонний (Київ, Україна),
А.В.Марченков (Москва, Росія) ² Моделювання даних ультразвукових сенсорів мобільного робота в середовищі з
перешкодами
O.N.Sukhoruchkina (Kyjiv, Ukraine), V.E.Prianichnikov (Moscow, Russia), N.V.Progonny (Kyjiv, Ukraine), A.V.Marchenkov (Moscow, Russia) ² Ultrasonic Sensors Simulation of Mobile Robot in the Environment with Obstacles.
Program application for simulating control system of intelligent multifunctional mobile truck robot as an example of computerized real functioning system is presented. Potentiality for estimation for model ultrasonic echolocation system data of mobile robot in environment with obstacles is shown.
È
VII. Нові інформаційні
технології та системи, які базуються на
обробленні
сигналів і зображень
Advanced Information
Technologies Based on
Signal/Image Processing and
Pattern Recognition
49.
Богдан Русин,
Володимир Остап, Олег Остап (Львів, Україна) ² Система ідентифікації особи за спектральними ознаками відбитків
Bohdan Rusyn, Volodymyr Ostap, Oleh Ostap (L’viv, Ukraine) ² Fingerprint Identification System Based on Spectral Features.
It is known a plenty of various
fingerprint identification systems in the world but the most part from them are
use recognition algorithms based on minutiae features. We had developed
complete fingerprint identification system, which includes the following
stages:
- fingerprint scanning;
- image preprocessing:
image segmentation; local normalization; quazioptimal filtration and ridge
frequency estimation; local orientation estimation; Gabor filter bank
filtration; histogram transformation;
- feature extraction:
singular point estimation; central area cutting; feature extraction;
- record in a database;
- recognition
(identification);
In
comparison with other systems, which use minutiae features (the fork, ending of
ridges), in it is used a correlation recognition method which allows to carry
out identification without any requirements to amount of minutiae features. In
addition, for improvement of recognition accuracy, together with known methods
of preprocessing we have developed and use a method of quasioptimal filtration,
and improved known.
50. Нюнькін К. М. (Донецьк, Україна) ² Можливість використання кольору при розпізнаванні жестів
Nyunkin K. M. (Donets’k, Ukraine) ² Opportunity of Use of Color at Gestures Recognition.
The problem of recognition of gestures in a sequence of color images is considered. In particular, the possibility of use of color for hands detection on images is investigated. The analysis of approaches to detection of the human faces on the basis of the information on color is executed. Restrictions of use for these purposes of human skin color are shown. It is recommended to apply saturated colors to giving of commands by gestures.
51.
В. В. Трипольська (Дніпропетровськ, Україна) ² Ідентифікація підписів за характерними ознаками перетворення Радону
V. V. Trypol’s’ka (Dnipropetrovs’k, Ukraine) ² Signatures Identification on Characteristic Features Based on Radon Transformation.
The method a neural network verification of the signatures as binary images in space informative features of the Radon transformation is submitted. The moment's invariant to rotation, transformation and scaling are used for construction of the vector features. The extraction of informative features is made in space Wavelet decomposition, pseudo-phase system of coordinates that on the one hand provides reliability of identification and on the other hand v decreases the size of space features. The method of extracting features and verificating the signatures is checked up on backpropagation neural network. The testing was carried out on the various signatures and the received results have confirmed reliability of the method.
52. В. О. Болтенков, М. В. Максимов, О. В. Маслов (Одеса, Україна) ² Принципи побудови діагностичних систем для АЕС на основі нових
інформаційних технологій
V. O. Boltenkov, M. V. Maksymov, O. V. Maslov (Odesa, Ukraine) ² The Fundamentals of Diagnostic Systems for NPP Architecture Based on Advanced Information Technologies.
The advanced diagnostic systems for NPP are discussed: leak diagnostic system and fuel diagnostic system. Leak diagnostic is based on two different principles — measuring activity of air surrounding place of leak (early leak detection) and detection of leak produced sound with distributed self-organizing net of microphone sensors (localization of leak place). Fuel burnup diagnostic is made by means of gamma-specrtometry with application of CdZnTe detectors. New technology to estimate fuel burnup in real time is proposed. The prospect of NPP diagnostic systems perfection is discussed.
53.
С. Г. Антощук, В. М. Крилов, В. О.
Давидов (Одеса, Україна) ² Автоматизована система ідентифікації та класифікації номерних знаків
S. H. Antoshchuk, V. M. Krylov, V. O. Davydov (Odesa, Ukraine) ² Automatic System for Vehicle Numbers Identification and Classification.
54. Рифа В. М., Баклан Я. І., Баклан І. В. (Київ, Україна) ² Метод головних компонент у задачах автентифікації
Ryfa V. M., Baklan Ja. I, Baklan I. V. (Kyjiv, Ukraine) ² Principal Components
Method to the Problem Authentification.
The possibility of application of the principal component method [1] is considered to the problem of authentication of an operator of a complex process control system [2]. The numerical experiment results show the convergence of eigenvalues of covariance matrixes to steady state values on the set of parameters with increasing number of measurements.
55. Валерій Пилипенко, Микола Сажок (Київ,
Україна)
² Озвучений редактор текстів для користувачів з вадами зору
Valery Pylypenko,
Mykola Sazhok (Kyjiv, Ukraine) ² Spoken Text
Editor for Blind Users.
The spoken editor for disabled is presented. There is provided both spoken feedback and input. Problems of interaction between computer and blind users are discussed.
56. О. В. Заболотний, О. І. Яворський (Львів, Україна) ² Виявлення основних ритмів регулярної складової електрокардіограми методом найменших квадратів
O. V. Zabolotnyj,
О. І. Javors’kyj (L’viv, Ukraine) ² Fundamental Rhythms
of Electrocardiogram Regular Components Discovering by the Least Squares
Method.
The almost periodically correlated random processes model is
applied for electrocardiogram signal description. The least squares method is adapted
for detection of fundamental rhythm of it mean components