Guided iterative back-projection scheme for single-image super-resolution

Hailiang Li, Kin Man Lam

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

6 Citations (Scopus)

Abstract

The iterative back-projection (IBP) technique is an effective approach to single-image super-resolution (SR) reconstruction. However, the conventional IBP-based SR method suffers from a jaggy and ringing effect around the edges, as the up-sampling process and the back-projection of the reconstructed image are isotropic. In this paper, we present an improved IBP approach, namely guided iterative back-projection (GIBP), which employs an effective edge-guided interpolation method for up-sampling, and either a joint bilateral filter or a guided filter is used to constrain the projection of the error coefficients so as to suppress the noise and the ringing artifacts (i.e. the halo effect). To up-scale an image with an arbitrary ratio, we also propose an edge-persevering bilinear up-scaling interpolation method in cooperation with GIBP. Experimental results validate that the proposed algorithm can significantly improve SR reconstruction accuracy, while suppressing the main artifacts, such as the ringing and jaggy effects, and the noise created in the traditional IBP methods.
Original languageEnglish
Title of host publication2013 IEEE Global High Tech Congress on Electronics, GHTCE 2013
PublisherIEEE Computer Society
Pages175-180
Number of pages6
DOIs
Publication statusPublished - 1 Jan 2013
Event2013 IEEE Global High Tech Congress on Electronics, GHTCE 2013 - Shenzhen, China
Duration: 17 Nov 201319 Nov 2013

Conference

Conference2013 IEEE Global High Tech Congress on Electronics, GHTCE 2013
CountryChina
CityShenzhen
Period17/11/1319/11/13

Keywords

  • edge preserving interpolation
  • guided image filter
  • Image super resolution
  • iterative back-projection
  • joint bilateral filter

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Cite this