Densely connected autoencoders for image compression

Song Zebang, Seiichiro Kamata

Research output: Contribution to conferencePaper

Abstract

Image compression, which is a type of data compression applied to digital images, has been a fundamental research topic for many decades. Recent image techniques produce very large amounts of data, which may make it prohibitive to storage and communications of image data without the use of compression. However, the traditional compression methods, such as JPEG, may introduce the compression artefact problems. Recently, deep learning has achieved great success in many computer vision tasks and is gradually being used in image compression. To solve the compression atrefact problem, in this paper, we present a lossy image compression architecture, which utilizes the advantages of the existing deep learning methods to achieve a high coding efficiency. We design a densely connected autoencoder structure for lossy image compression. Firstly, we design a densely autoencoder structure to get richer feature information from image which can be helpful for compression. Secondly, we design a U-net like network to decrease the distortion caused by compression. Finally, an improved binarizer is adopted to quantize the output of encoder. In low bit rate image compression, experiments show that our method significantly outperforms JPEG and JPEG2000 and can produce a better visual result with sharp edges, rich textures, and fewer artifacts.

Original languageEnglish
Pages78-83
Number of pages6
DOIs
Publication statusPublished - 2019 Jan 1
Event2nd International Conference on Image and Graphics Processing, ICIGP 2019 - Singapore, Singapore
Duration: 2019 Feb 232019 Feb 25

Conference

Conference2nd International Conference on Image and Graphics Processing, ICIGP 2019
CountrySingapore
CitySingapore
Period19/2/2319/2/25

Fingerprint

Image compression
Data compression
Computer vision
Textures
Communication
Experiments
Deep learning

Keywords

  • Convolutional neural networks (CNNs)
  • Densely AutoEncoders
  • Lossy image compression
  • U-net like structure

ASJC Scopus subject areas

  • Human-Computer Interaction
  • Computer Networks and Communications
  • Computer Vision and Pattern Recognition
  • Software

Cite this

Zebang, S., & Kamata, S. (2019). Densely connected autoencoders for image compression. 78-83. Paper presented at 2nd International Conference on Image and Graphics Processing, ICIGP 2019, Singapore, Singapore. https://doi.org/10.1145/3313950.3313965

Densely connected autoencoders for image compression. / Zebang, Song; Kamata, Seiichiro.

2019. 78-83 Paper presented at 2nd International Conference on Image and Graphics Processing, ICIGP 2019, Singapore, Singapore.

Research output: Contribution to conferencePaper

Zebang, S & Kamata, S 2019, 'Densely connected autoencoders for image compression' Paper presented at 2nd International Conference on Image and Graphics Processing, ICIGP 2019, Singapore, Singapore, 19/2/23 - 19/2/25, pp. 78-83. https://doi.org/10.1145/3313950.3313965
Zebang S, Kamata S. Densely connected autoencoders for image compression. 2019. Paper presented at 2nd International Conference on Image and Graphics Processing, ICIGP 2019, Singapore, Singapore. https://doi.org/10.1145/3313950.3313965
Zebang, Song ; Kamata, Seiichiro. / Densely connected autoencoders for image compression. Paper presented at 2nd International Conference on Image and Graphics Processing, ICIGP 2019, Singapore, Singapore.6 p.
@conference{a6780dfbba444ca1afd1627b595964e1,
title = "Densely connected autoencoders for image compression",
abstract = "Image compression, which is a type of data compression applied to digital images, has been a fundamental research topic for many decades. Recent image techniques produce very large amounts of data, which may make it prohibitive to storage and communications of image data without the use of compression. However, the traditional compression methods, such as JPEG, may introduce the compression artefact problems. Recently, deep learning has achieved great success in many computer vision tasks and is gradually being used in image compression. To solve the compression atrefact problem, in this paper, we present a lossy image compression architecture, which utilizes the advantages of the existing deep learning methods to achieve a high coding efficiency. We design a densely connected autoencoder structure for lossy image compression. Firstly, we design a densely autoencoder structure to get richer feature information from image which can be helpful for compression. Secondly, we design a U-net like network to decrease the distortion caused by compression. Finally, an improved binarizer is adopted to quantize the output of encoder. In low bit rate image compression, experiments show that our method significantly outperforms JPEG and JPEG2000 and can produce a better visual result with sharp edges, rich textures, and fewer artifacts.",
keywords = "Convolutional neural networks (CNNs), Densely AutoEncoders, Lossy image compression, U-net like structure",
author = "Song Zebang and Seiichiro Kamata",
year = "2019",
month = "1",
day = "1",
doi = "10.1145/3313950.3313965",
language = "English",
pages = "78--83",
note = "2nd International Conference on Image and Graphics Processing, ICIGP 2019 ; Conference date: 23-02-2019 Through 25-02-2019",

}

TY - CONF

T1 - Densely connected autoencoders for image compression

AU - Zebang, Song

AU - Kamata, Seiichiro

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Image compression, which is a type of data compression applied to digital images, has been a fundamental research topic for many decades. Recent image techniques produce very large amounts of data, which may make it prohibitive to storage and communications of image data without the use of compression. However, the traditional compression methods, such as JPEG, may introduce the compression artefact problems. Recently, deep learning has achieved great success in many computer vision tasks and is gradually being used in image compression. To solve the compression atrefact problem, in this paper, we present a lossy image compression architecture, which utilizes the advantages of the existing deep learning methods to achieve a high coding efficiency. We design a densely connected autoencoder structure for lossy image compression. Firstly, we design a densely autoencoder structure to get richer feature information from image which can be helpful for compression. Secondly, we design a U-net like network to decrease the distortion caused by compression. Finally, an improved binarizer is adopted to quantize the output of encoder. In low bit rate image compression, experiments show that our method significantly outperforms JPEG and JPEG2000 and can produce a better visual result with sharp edges, rich textures, and fewer artifacts.

AB - Image compression, which is a type of data compression applied to digital images, has been a fundamental research topic for many decades. Recent image techniques produce very large amounts of data, which may make it prohibitive to storage and communications of image data without the use of compression. However, the traditional compression methods, such as JPEG, may introduce the compression artefact problems. Recently, deep learning has achieved great success in many computer vision tasks and is gradually being used in image compression. To solve the compression atrefact problem, in this paper, we present a lossy image compression architecture, which utilizes the advantages of the existing deep learning methods to achieve a high coding efficiency. We design a densely connected autoencoder structure for lossy image compression. Firstly, we design a densely autoencoder structure to get richer feature information from image which can be helpful for compression. Secondly, we design a U-net like network to decrease the distortion caused by compression. Finally, an improved binarizer is adopted to quantize the output of encoder. In low bit rate image compression, experiments show that our method significantly outperforms JPEG and JPEG2000 and can produce a better visual result with sharp edges, rich textures, and fewer artifacts.

KW - Convolutional neural networks (CNNs)

KW - Densely AutoEncoders

KW - Lossy image compression

KW - U-net like structure

UR - http://www.scopus.com/inward/record.url?scp=85065769924&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85065769924&partnerID=8YFLogxK

U2 - 10.1145/3313950.3313965

DO - 10.1145/3313950.3313965

M3 - Paper

SP - 78

EP - 83

ER -