Journal influence
Higher Attestation Commission (VAK) - К1 quartile
Russian Science Citation Index (RSCI)
Bookmark
Next issue
№3
Publication date:
16 September 2025
Research on compression of raster images using artificial neural networks
Date of submission article: 25.04.2018
UDC: 004.032.2
The article was published in issue no. № 3, 2018 [ pp. 430-434 ]Abstract:Modern rates of information growth stored on hard disks transferred over the Internet and local enterprise networks has made it necessary to solve the problem of compressing, transferring and storing data. Most of the transferred data is in the form of multimedia content. Nowadays, the algorithms for compressing visual information based on the neural network apparatus are becoming more popular. Unlike classical algorithms, which are based on the elimination of redundancy, these algorithms are based on artificial neural networks. The field is relevant due to the development of mathematical algorithms for network learning, which will improve existing compression methods in the future. The analysis of publications showed that nowadays there is no particular information about the influence of the artificial neural network architecture on a learning process and the quality of their work in real multimedia content. The important task is to select a network topology, which is most suitable for compressing visual information. The purpose of the article is to describe the capabilities of one of the types of artificial neural networks called a multi-layer perceptron in the area of compression and recovery of images of an arbitrary type. The paper analyzes topologies of artificial neural networks, algorithms for their learning, and the efficiency of their work. It also describes the architecture of a “bottleneck”, which is most often used in solving the problem of image compression and recovery. The authors give one of the ways of encoding and decoding data obtained during network operation. The paper describes a computational experiment and gives its results. The experiment showed that using a multilayer perceptron with an input vector of more than eight values turned out to be less effective. As a result, the authors propose the most suitable network architecture to use in practice.
Аннотация:Современные темпы роста объемов информации, хранящихся на жестких дисках, передаваемых по сети Интернет и локальным сетям предприятий, обусловили актуальность задачи сжатия, передачи и хранения данных. Большая часть передаваемых по сети данных представлена в виде мультимедийного контента. Сегодня все более популярными становятся алгоритмы сжатия визуальной информации, основанные на нейросетевом аппарате. В отличие от классических алгоритмов, основанных на устранении избыточности, данные алгоритмы базируются на искусственных нейронных сетях. Направление актуально в связи с развитием математических алгоритмов обучения сети, что в дальнейшем позволит улучшить существующие методы сжатия. Проведенный анализ публикаций показал, что в настоящее время конкретная информация о влиянии архитектуры искусственной нейронной сети на процесс обучения и качество их работы на реальном мультимедийном контенте отсутствует. Актуальна задача выбора топологии сети, наиболее подходящей для сжатия визуальной информации. Целью авторов статьи является описание возможностей одного из типов искусственных нейронных сетей – многослойного персептрона – в области сжатия и восстановления изображений произвольного вида. Рассматриваются топологии искусственных нейронных сетей и алгоритмы их обучения, анализируется эффективность работы этих сетей. Описывается архитектура бутылочного горлышка, наиболее часто используемая при решении задачи сжатия и восстановления изображений. Приводится один из способов кодирования и декодирования данных, полученных при работе сетей. В статье описывается проведенный вычислительный эксперимент, приведены полученные результаты. Результаты показали, что использование многослойного персептрона с входным вектором свыше восьми значений менее эффективно. В итоге предложена наиболее подходящая архитектура сети, которую можно использовать на практике.
Authors: A.A. Genov (vlad_osipovv@mail.ru) - Center of Visualization and Satellite Information Technologies SRISA (Professor, Leading Researcher), Moscow, Russia, Ph.D, K.D. Rusakov (rusakov.msk@yandex.ru) - V.A. Trapeznikov Institute of Control Sciences of RAS (Junior Researcher), Moscow, Russia, A.A. Moiseev (moisandry@gmail.com) - Bauman Moscow State Technical University (Student), Moscow, Russia, V.V. Osipov (vlad_osipovv@mail.ru) - Center of Visualization and Satellite Information Technologies SRISA (Associate Professor, Senior Researcher), Moscow, Russia, Ph.D | |
Keywords: neural network, compression algorithm, image, machine learning |
|
Page views: 16340 |
PDF version article Full issue in PDF (29.03Mb) |
Исследование сжатия растровых изображений с использованием искусственных нейронных сетей
DOI: 10.15827/0236-235X.123.430-434
Date of submission article: 25.04.2018
UDC: 004.032.2
The article was published in issue no. № 3, 2018. [ pp. 430-434 ]
Modern rates of information growth stored on hard disks transferred over the Internet and local enterprise networks has made it necessary to solve the problem of compressing, transferring and storing data. Most of the transferred data is in the form of multimedia content.
Nowadays, the algorithms for compressing visual information based on the neural network apparatus are becoming more popular. Unlike classical algorithms, which are based on the elimination of redundancy, these algorithms are based on artificial neural networks. The field is relevant due to the development of mathematical algorithms for network learning, which will improve existing compression methods in the future.
The analysis of publications showed that nowadays there is no particular information about the influence of the artificial neural network architecture on a learning process and the quality of their work in real multimedia content. The important task is to select a network topology, which is most suitable for compressing visual information.
The purpose of the article is to describe the capabilities of one of the types of artificial neural networks called a multi-layer perceptron in the area of compression and recovery of images of an arbitrary type. The paper analyzes topologies of artificial neural networks, algorithms for their learning, and the efficiency of their work. It also describes the architecture of a “bottleneck”, which is most often used in solving the problem of image compression and recovery. The authors give one of the ways of encoding and decoding data obtained during network operation. The paper describes a computational experiment and gives its results.
The experiment showed that using a multilayer perceptron with an input vector of more than eight values turned out to be less effective. As a result, the authors propose the most suitable network architecture to use in practice.
A.A. Genov (vlad_osipovv@mail.ru) - Center of Visualization and Satellite Information Technologies SRISA (Professor, Leading Researcher), Moscow, Russia, Ph.D, K.D. Rusakov (rusakov.msk@yandex.ru) - V.A. Trapeznikov Institute of Control Sciences of RAS (Junior Researcher), Moscow, Russia, A.A. Moiseev (moisandry@gmail.com) - Bauman Moscow State Technical University (Student), Moscow, Russia, V.V. Osipov (vlad_osipovv@mail.ru) - Center of Visualization and Satellite Information Technologies SRISA (Associate Professor, Senior Researcher), Moscow, Russia, Ph.D
Ссылка скопирована!
Permanent link: http://swsys.ru/index.php?page=article&id=4481&lang=&lang=en&like=1 |
Print version Full issue in PDF (29.03Mb) |
The article was published in issue no. № 3, 2018 [ pp. 430-434 ] |
The article was published in issue no. № 3, 2018. [ pp. 430-434 ]
Perhaps, you might be interested in the following articles of similar topics:Perhaps, you might be interested in the following articles of similar topics:
- Обнаружение аномалий сетевого трафика методом глубокого обучения
- Нейросетевой поведенческий анализ действий пользователя в целях обнаружения вторжений уровня узла
- Извлечение аспектов из текстов научных статей
- Генетический алгоритм выбора доминантных признаков для нейронной сети
- Прогнозирование технологических тенденций на основе анализа разнородных данных
Back to the list of articles