Advanced search
1 file | 2.33 MB

Evaluation of color differences in natural scene color images

Benhur Ortiz Jaramillo (UGent) , Asli Kumcu (UGent) , Ljiljana Platisa (UGent) and Wilfried Philips (UGent)
Author
Organization
Abstract
Since there is a wide range of applications requiring image color difference (CD) assessment (e.g. color quantization, color mapping), a number of CD measures for images have been proposed. However, the performance evaluation of such measures often suffers from the following major flaws: (1) test images contain primarily spatial- (e.g. blur) rather than color-specific distortions (e.g. quantization noise), (2) there are too few test images (lack of variability in color content), and (3) test images are not publicly available (difficult to reproduce and compare). Accordingly, the performance of CD measures reported in the state-of-the-art is ambiguous and therefore inconclusive to be used for any specific color-related application. In this work, we review a total of twenty four state-of-the-art CD measures. Then, based on the findings of our review, we propose a novel method to compute CDs in natural scene color images. We have tested our measure as well as the state-of-the-art measures on three color related distortions from a publicly available database (mean shift, change in color saturation and quantization noise). Our experimental results show that the correlation between the subjective scores and the proposed measure exceeds 85% which is better than the other twenty four CD measures tested in this work (for illustration the best performing state-of-the-art CD measures achieve correlations with humans lower than 80%).
Keywords
Signal Processing, Electrical and Electronic Engineering, Software, Computer Vision and Pattern Recognition

Downloads

  • (...).pdf
    • full text
    • |
    • UGent only
    • |
    • PDF
    • |
    • 2.33 MB

Citation

Please use this url to cite or link to this publication:

Chicago
Ortiz Jaramillo, Benhur, Asli Kumcu, Ljiljana Platisa, and Wilfried Philips. 2019. “Evaluation of Color Differences in Natural Scene Color Images.” Ed. F. Dufaux. Signal Processing: Image Communication 71: 128–137.
APA
Ortiz Jaramillo, B., Kumcu, A., Platisa, L., & Philips, W. (2019). Evaluation of color differences in natural scene color images. (F. Dufaux, Ed.)Signal Processing: Image Communication, 71, 128–137.
Vancouver
1.
Ortiz Jaramillo B, Kumcu A, Platisa L, Philips W. Evaluation of color differences in natural scene color images. Dufaux F, editor. Signal Processing: Image Communication. Elsevier BV; 2019;71:128–37.
MLA
Ortiz Jaramillo, Benhur, Asli Kumcu, Ljiljana Platisa, et al. “Evaluation of Color Differences in Natural Scene Color Images.” Ed. F. Dufaux. Signal Processing: Image Communication 71 (2019): 128–137. Print.
@article{8586710,
  abstract     = {Since there is a wide range of applications requiring image color difference (CD) assessment (e.g. color quantization, color mapping), a number of CD measures for images have been proposed. However, the performance evaluation of such measures often suffers from the following major flaws: (1) test images contain primarily spatial- (e.g. blur) rather than color-specific distortions (e.g. quantization noise), (2) there are too few test images (lack of variability in color content), and (3) test images are not publicly available (difficult to reproduce and compare). Accordingly, the performance of CD measures reported in the state-of-the-art is ambiguous and therefore inconclusive to be used for any specific color-related application.

In this work, we review a total of twenty four state-of-the-art CD measures. Then, based on the findings of our review, we propose a novel method to compute CDs in natural scene color images. We have tested our measure as well as the state-of-the-art measures on three color related distortions from a publicly available database (mean shift, change in color saturation and quantization noise). Our experimental results show that the correlation between the subjective scores and the proposed measure exceeds 85\% which is better than the other twenty four CD measures tested in this work (for illustration the best performing state-of-the-art CD measures achieve correlations with humans lower than 80\%).},
  author       = {Ortiz Jaramillo, Benhur and Kumcu, Asli and Platisa, Ljiljana and Philips, Wilfried},
  editor       = {Dufaux, F.},
  issn         = {0923-5965},
  journal      = {Signal Processing: Image Communication},
  language     = {eng},
  pages        = {128--137},
  publisher    = {Elsevier BV},
  title        = {Evaluation of color differences in natural scene color images},
  url          = {http://dx.doi.org/10.1016/j.image.2018.11.009},
  volume       = {71},
  year         = {2019},
}

Altmetric
View in Altmetric