Image deblurring using scale-recurrent network for mobile devices

Indra Pambudi, Dina Chahyati

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

Abstract

Image deblurring is a problem in computer vision that aims to restore blur images into sharp images. The blurring might be caused by the camera shaking or an object moving when the image is captured, resulting in an image with a non-uniform blur in a dynamic scene. One recent approach to restoring images with non-uniform blur is by using end-to-end deep neural networks. Continuing the deblur research using a scale-recurrent network, we modify the neural network architecture to be lighter to run on mobile devices. The proposed method achieves PSNR of 29.55 and SSIM of 0.8873 in a 16.9 MB sized model. The inference process on a mobile device only requires 1 GB of memory with 8.2 seconds in latency for deblurring a single 1280x720 pixel image.

Original languageEnglish
Title of host publication2019 International Conference on Advanced Computer Science and Information Systems, ICACSIS 2019
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages145-150
Number of pages6
ISBN (Electronic)9781728152929
DOIs
Publication statusPublished - Oct 2019
Event11th International Conference on Advanced Computer Science and Information Systems, ICACSIS 2019 - Bali, Indonesia
Duration: 12 Oct 201913 Oct 2019

Publication series

Name2019 International Conference on Advanced Computer Science and Information Systems, ICACSIS 2019

Conference

Conference11th International Conference on Advanced Computer Science and Information Systems, ICACSIS 2019
CountryIndonesia
CityBali
Period12/10/1913/10/19

Keywords

  • Image deblurring
  • Inference on mobile device
  • Scale-recurrent network

Fingerprint Dive into the research topics of 'Image deblurring using scale-recurrent network for mobile devices'. Together they form a unique fingerprint.

Cite this