No-reference image quality assessment in curvelet domain

Lixiong Liu*, Hongping Dong, Hua Huang, Alan C. Bovik

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

192 Citations (Scopus)

Abstract

We study the efficacy of utilizing a powerful image descriptor, the curvelet transform, to learn a no-reference (NR) image quality assessment (IQA) model. A set of statistical features are extracted from a computed image curvelet representation, including the coordinates of the maxima of the log-histograms of the curvelet coefficients values, and the energy distributions of both orientation and scale in the curvelet domain. Our results indicate that these features are sensitive to the presence and severity of image distortion. Operating within a 2-stage framework of distortion classification followed by quality assessment, we train an image distortion and quality prediction engine using a support vector machine (SVM). The resulting algorithm, dubbed CurveletQA for short, was tested on the LIVE IQA database and compared to state-of-the-art NR/FR IQA algorithms. We found that CurveletQA correlates well with human subjective opinions of image quality, delivering performance that is competitive with popular full-reference (FR) IQA algorithms such as SSIM, and with top-performing NR IQA models. At the same time, CurveletQA has a relatively low complexity.

Original languageEnglish
Pages (from-to)494-505
Number of pages12
JournalSignal Processing: Image Communication
Volume29
Issue number4
DOIs
Publication statusPublished - Apr 2014

Keywords

  • Curvelet
  • Image quality assessment (IQA)
  • Natural scene statistics (NSS)
  • No reference (NR)
  • Support Vector Machine (SVM)

Fingerprint

Dive into the research topics of 'No-reference image quality assessment in curvelet domain'. Together they form a unique fingerprint.

Cite this