Low-Complexity Neural Network Design for Super-Resolution Imaging

Wei Ting Chen, Shao Chieh Liao, Pei Yin Chen, Hsin Yu Shih

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

Abstract

Super-resolution imaging is a widely used technology in consumer electronics, such as 4K TV and portable media players. The convolution neural network (CNN) has confirmed the high quality for super-resolution (SR) imaging. However, in order to pursue better imaging quality and higher PSNR/SSIM, recent researches usually involve high complexity architecture and lead to a large amount of memory usage. To overcome these drawbacks, we propose a low-complexity neural network architecture, which has a significant reduction in resource consumption. Furthermore, it also can serve 2X, 3X, and 4X SR images at the same time with a single architecture. Compared with similar less complicated architectures [1]-[2], the results show that our architecture can achieve the best imaging quality (PSNR and SSIM) with the lowest parameter usage.

Original languageEnglish
Title of host publication2020 IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728173993
DOIs
Publication statusPublished - 2020 Sep 28
Event7th IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2020 - Taoyuan, Taiwan
Duration: 2020 Sep 282020 Sep 30

Publication series

Name2020 IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2020

Conference

Conference7th IEEE International Conference on Consumer Electronics - Taiwan, ICCE-Taiwan 2020
CountryTaiwan
CityTaoyuan
Period20-09-2820-09-30

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Artificial Intelligence
  • Computer Science Applications
  • Signal Processing
  • Electrical and Electronic Engineering
  • Instrumentation

Fingerprint Dive into the research topics of 'Low-Complexity Neural Network Design for Super-Resolution Imaging'. Together they form a unique fingerprint.

Cite this