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

30 Scopus citations


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
Number of pages11
ISBN (Print)9783030041663
StatePublished - 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


Other25th International Conference on Neural Information Processing, ICONIP 2018
CitySiem Reap


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

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'DeepSIC: Deep semantic image compression'. Together they form a unique fingerprint.

Cite this