DeepSIC: Deep semantic image compression

Sihui Luo, Yezhou Yang, Yanling Yin, Chengchao Shen, Ya Zhao, Mingli Song

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

Incorporating semantic analysis into image compression can significantly reduce the repetitive computation of fundamental semantic analysis in client-side applications such as semantic image retrieval. The same practice also enables the compressed code to carry semantic information of the image during its storage and transmission. In this paper, we propose a Deep Semantic Image Compression (DeepSIC) model to achieve this goal and put forward two novel architectures that aim to reconstruct the compressed image and generate corresponding semantic representations at the same time by a single end-to-end optimized network. The first architecture performs semantic analysis in the encoding process by reserving a portion of the bits from the compressed code to store the semantic representations. The second performs semantic analysis in the decoding step with the feature maps that are embedded in the compressed code. In both architectures, the feature maps are shared by the compression and the semantic analytics modules. Experiments over benchmarking datasets show promising performance of the proposed compression model.

Original languageEnglish (US)
Title of host publicationNeural Information Processing - 25th International Conference, ICONIP 2018, Proceedings
EditorsLong Cheng, Andrew Chi Sing Leung, Seiichi Ozawa
PublisherSpringer Verlag
Pages96-106
Number of pages11
ISBN (Print)9783030041663
DOIs
StatePublished - Jan 1 2018
Event25th International Conference on Neural Information Processing, ICONIP 2018 - Siem Reap, Cambodia
Duration: Dec 13 2018Dec 16 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11301 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other25th International Conference on Neural Information Processing, ICONIP 2018
CountryCambodia
CitySiem Reap
Period12/13/1812/16/18

Fingerprint

Image Compression
Image compression
Semantic Analysis
Semantics
Compression
Image Retrieval
Benchmarking
Decoding
Encoding
Image retrieval
Module
Model
Experiment
Architecture

Keywords

  • Deep image compression
  • End-to-end optimization
  • Semantic image compression

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Luo, S., Yang, Y., Yin, Y., Shen, C., Zhao, Y., & Song, M. (2018). DeepSIC: Deep semantic image compression. In L. Cheng, A. C. S. Leung, & S. Ozawa (Eds.), Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings (pp. 96-106). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11301 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-04167-0_9

DeepSIC : Deep semantic image compression. / Luo, Sihui; Yang, Yezhou; Yin, Yanling; Shen, Chengchao; Zhao, Ya; Song, Mingli.

Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. ed. / Long Cheng; Andrew Chi Sing Leung; Seiichi Ozawa. Springer Verlag, 2018. p. 96-106 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11301 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Luo, S, Yang, Y, Yin, Y, Shen, C, Zhao, Y & Song, M 2018, DeepSIC: Deep semantic image compression. in L Cheng, ACS Leung & S Ozawa (eds), Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11301 LNCS, Springer Verlag, pp. 96-106, 25th International Conference on Neural Information Processing, ICONIP 2018, Siem Reap, Cambodia, 12/13/18. https://doi.org/10.1007/978-3-030-04167-0_9
Luo S, Yang Y, Yin Y, Shen C, Zhao Y, Song M. DeepSIC: Deep semantic image compression. In Cheng L, Leung ACS, Ozawa S, editors, Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. Springer Verlag. 2018. p. 96-106. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-04167-0_9
Luo, Sihui ; Yang, Yezhou ; Yin, Yanling ; Shen, Chengchao ; Zhao, Ya ; Song, Mingli. / DeepSIC : Deep semantic image compression. Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings. editor / Long Cheng ; Andrew Chi Sing Leung ; Seiichi Ozawa. Springer Verlag, 2018. pp. 96-106 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{fea6f71050004c8c8c495cb4ad9280e7,
title = "DeepSIC: Deep semantic image compression",
abstract = "Incorporating semantic analysis into image compression can significantly reduce the repetitive computation of fundamental semantic analysis in client-side applications such as semantic image retrieval. The same practice also enables the compressed code to carry semantic information of the image during its storage and transmission. In this paper, we propose a Deep Semantic Image Compression (DeepSIC) model to achieve this goal and put forward two novel architectures that aim to reconstruct the compressed image and generate corresponding semantic representations at the same time by a single end-to-end optimized network. The first architecture performs semantic analysis in the encoding process by reserving a portion of the bits from the compressed code to store the semantic representations. The second performs semantic analysis in the decoding step with the feature maps that are embedded in the compressed code. In both architectures, the feature maps are shared by the compression and the semantic analytics modules. Experiments over benchmarking datasets show promising performance of the proposed compression model.",
keywords = "Deep image compression, End-to-end optimization, Semantic image compression",
author = "Sihui Luo and Yezhou Yang and Yanling Yin and Chengchao Shen and Ya Zhao and Mingli Song",
year = "2018",
month = "1",
day = "1",
doi = "10.1007/978-3-030-04167-0_9",
language = "English (US)",
isbn = "9783030041663",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "96--106",
editor = "Long Cheng and Leung, {Andrew Chi Sing} and Seiichi Ozawa",
booktitle = "Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings",

}

TY - GEN

T1 - DeepSIC

T2 - Deep semantic image compression

AU - Luo, Sihui

AU - Yang, Yezhou

AU - Yin, Yanling

AU - Shen, Chengchao

AU - Zhao, Ya

AU - Song, Mingli

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Incorporating semantic analysis into image compression can significantly reduce the repetitive computation of fundamental semantic analysis in client-side applications such as semantic image retrieval. The same practice also enables the compressed code to carry semantic information of the image during its storage and transmission. In this paper, we propose a Deep Semantic Image Compression (DeepSIC) model to achieve this goal and put forward two novel architectures that aim to reconstruct the compressed image and generate corresponding semantic representations at the same time by a single end-to-end optimized network. The first architecture performs semantic analysis in the encoding process by reserving a portion of the bits from the compressed code to store the semantic representations. The second performs semantic analysis in the decoding step with the feature maps that are embedded in the compressed code. In both architectures, the feature maps are shared by the compression and the semantic analytics modules. Experiments over benchmarking datasets show promising performance of the proposed compression model.

AB - Incorporating semantic analysis into image compression can significantly reduce the repetitive computation of fundamental semantic analysis in client-side applications such as semantic image retrieval. The same practice also enables the compressed code to carry semantic information of the image during its storage and transmission. In this paper, we propose a Deep Semantic Image Compression (DeepSIC) model to achieve this goal and put forward two novel architectures that aim to reconstruct the compressed image and generate corresponding semantic representations at the same time by a single end-to-end optimized network. The first architecture performs semantic analysis in the encoding process by reserving a portion of the bits from the compressed code to store the semantic representations. The second performs semantic analysis in the decoding step with the feature maps that are embedded in the compressed code. In both architectures, the feature maps are shared by the compression and the semantic analytics modules. Experiments over benchmarking datasets show promising performance of the proposed compression model.

KW - Deep image compression

KW - End-to-end optimization

KW - Semantic image compression

UR - http://www.scopus.com/inward/record.url?scp=85059074383&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85059074383&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-04167-0_9

DO - 10.1007/978-3-030-04167-0_9

M3 - Conference contribution

AN - SCOPUS:85059074383

SN - 9783030041663

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 96

EP - 106

BT - Neural Information Processing - 25th International Conference, ICONIP 2018, Proceedings

A2 - Cheng, Long

A2 - Leung, Andrew Chi Sing

A2 - Ozawa, Seiichi

PB - Springer Verlag

ER -