Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improved image display by identifying the RGB family color space

Elvis Togban, Djemel Ziou

TL;DR

This work addresses the problem of identifying the encoding color space within the RGB family when image metadata is missing or corrupted. It introduces a pixel-embedding model paired with a Gaussian-process framework to capture channel interactions, creating intra-channel and inter-channel feature vectors that feed a multinomial logistic regression classifier over the five target spaces (Adobe RGB, Apple RGB, ColorMatch RGB, ProPhoto RGB, and sRGB). Parameter estimation is performed with a two-level EM algorithm: first to fit the embedding residuals with Gaussian or Gaussian-mixture noise, and second to estimate Gaussian-mixture components, with model order chosen by AIC. The approach achieves roughly 68% accuracy, outperforming a gamut-based baseline around 20%, and demonstrates that sRGB is easiest to identify while ColorMatch RGB is hardest; this enables improved display fidelity when color-space metadata is unavailable, with room for improvement via more flexible distributions and potential image-quality cues.

Abstract

To display an image, the color space in which the image is encoded is assumed to be known. Unfortunately, this assumption is rarely realistic. In this paper, we propose to identify the color space of a given color image using pixel embedding and the Gaussian process. Five color spaces are supported, namely Adobe RGB, Apple RGB, ColorMatch RGB, ProPhoto RGB and sRGB. The results obtained show that this problem deserves more efforts.

Improved image display by identifying the RGB family color space

TL;DR

This work addresses the problem of identifying the encoding color space within the RGB family when image metadata is missing or corrupted. It introduces a pixel-embedding model paired with a Gaussian-process framework to capture channel interactions, creating intra-channel and inter-channel feature vectors that feed a multinomial logistic regression classifier over the five target spaces (Adobe RGB, Apple RGB, ColorMatch RGB, ProPhoto RGB, and sRGB). Parameter estimation is performed with a two-level EM algorithm: first to fit the embedding residuals with Gaussian or Gaussian-mixture noise, and second to estimate Gaussian-mixture components, with model order chosen by AIC. The approach achieves roughly 68% accuracy, outperforming a gamut-based baseline around 20%, and demonstrates that sRGB is easiest to identify while ColorMatch RGB is hardest; this enables improved display fidelity when color-space metadata is unavailable, with room for improvement via more flexible distributions and potential image-quality cues.

Abstract

To display an image, the color space in which the image is encoded is assumed to be known. Unfortunately, this assumption is rarely realistic. In this paper, we propose to identify the color space of a given color image using pixel embedding and the Gaussian process. Five color spaces are supported, namely Adobe RGB, Apple RGB, ColorMatch RGB, ProPhoto RGB and sRGB. The results obtained show that this problem deserves more efforts.
Paper Structure (5 sections, 18 equations, 2 figures, 1 table, 1 algorithm)

This paper contains 5 sections, 18 equations, 2 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Color space identification model
  3. Parameters estimation
  4. Experimental Results
  5. Conclusion

Figures (2)

  • Figure 1: Images displayed with Firefox browser. In the first two rows, the color spaces used for encoding the images are unknown, while those of the images in the last row are known.
  • Figure 2: (a) A RGB image. (b) Histogram of the errors $r(m,n)$ computed for the red channel of the image displayed in (a).