Skip to main navigation Skip to search Skip to main content

Restoration of Compressed Picture Based on Lightweight Convolutional Neural Network

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

3 Scopus citations

Abstract

In this work, we propose a deep learning-based method to improve the quality of JPEG images. Our proposed network predicts the compression loss of the JPEG image for compensating and restoring the image quality. To solve the color bleeding artifacts often found in JPEG image, our network considering it in our model and objective functions to restore the color channels. Our network is much lighter by using fewer parameters compared to other work, whereas our method can still provide satisfactory and well-restored images for JPEG images as demonstrated in our experiments. Even with the additional handling on the color channels, the number of parameters in our network model is still kept low around 224k.

Original languageEnglish
Title of host publicationProceedings - 2019 International Symposium on Intelligent Signal Processing and Communication Systems, ISPACS 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728130385
DOIs
StatePublished - Dec 2019
Event2019 International Symposium on Intelligent Signal Processing and Communication Systems, ISPACS 2019 - Taipei, Taiwan
Duration: 3 Dec 20196 Dec 2019

Publication series

NameProceedings - 2019 International Symposium on Intelligent Signal Processing and Communication Systems, ISPACS 2019

Conference

Conference2019 International Symposium on Intelligent Signal Processing and Communication Systems, ISPACS 2019
Country/TerritoryTaiwan
CityTaipei
Period3/12/196/12/19

Bibliographical note

Publisher Copyright:
© 2019 IEEE.

Keywords

  • JPEG
  • compression artifacts removal
  • convolutional neural network
  • dilated convolution
  • residual network

Fingerprint

Dive into the research topics of 'Restoration of Compressed Picture Based on Lightweight Convolutional Neural Network'. Together they form a unique fingerprint.

Cite this