Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Study on the Effect of Color Spaces in Learned Image Compression

Srivatsa Prativadibhayankaram, Mahadev Prasad Panda, Jürgen Seiler, Thomas Richter, Heiko Sparenberg, Siegfried Fößel, André Kaup

TL;DR

The paper investigates the impact of color spaces on learned image compression using the SLIC framework, introducing RGB, YUV, and LAB variants with either a two-branch luminance/chrominance design or a single-branch RGB design. It trains models with a rate-distortion objective that balances rate, MSE, MS-SSIM, and color difference, across multiple operating points, and compares their performance to state-of-the-art codecs on standard RD metrics. The main finding is that the RGB single-branch variant delivers the strongest RD performance, particularly in MS-SSIM and CIEDE2000 with notable BD-BR gains, but at a higher model complexity; YUV and LAB show similar performance with lower resource needs. The work analyzes chroma-channel capacity and latent-channel impulse responses to show how explicit luminance/chrominance separation enables targeted optimization, while end-to-end training still yields strong results for RGB. These insights guide color-space choices in learned compression and motivate exploring additional color spaces like HSV or XYZ.

Abstract

In this work, we present a comparison between color spaces namely YUV, LAB, RGB and their effect on learned image compression. For this we use the structure and color based learned image codec (SLIC) from our prior work, which consists of two branches - one for the luminance component (Y or L) and another for chrominance components (UV or AB). However, for the RGB variant we input all 3 channels in a single branch, similar to most learned image codecs operating in RGB. The models are trained for multiple bitrate configurations in each color space. We report the findings from our experiments by evaluating them on various datasets and compare the results to state-of-the-art image codecs. The YUV model performs better than the LAB variant in terms of MS-SSIM with a Bjøntegaard delta bitrate (BD-BR) gain of 7.5\% using VTM intra-coding mode as the baseline. Whereas the LAB variant has a better performance than YUV model in terms of CIEDE2000 having a BD-BR gain of 8\%. Overall, the RGB variant of SLIC achieves the best performance with a BD-BR gain of 13.14\% in terms of MS-SSIM and a gain of 17.96\% in CIEDE2000 at the cost of a higher model complexity.

A Study on the Effect of Color Spaces in Learned Image Compression

TL;DR

The paper investigates the impact of color spaces on learned image compression using the SLIC framework, introducing RGB, YUV, and LAB variants with either a two-branch luminance/chrominance design or a single-branch RGB design. It trains models with a rate-distortion objective that balances rate, MSE, MS-SSIM, and color difference, across multiple operating points, and compares their performance to state-of-the-art codecs on standard RD metrics. The main finding is that the RGB single-branch variant delivers the strongest RD performance, particularly in MS-SSIM and CIEDE2000 with notable BD-BR gains, but at a higher model complexity; YUV and LAB show similar performance with lower resource needs. The work analyzes chroma-channel capacity and latent-channel impulse responses to show how explicit luminance/chrominance separation enables targeted optimization, while end-to-end training still yields strong results for RGB. These insights guide color-space choices in learned compression and motivate exploring additional color spaces like HSV or XYZ.

Abstract

In this work, we present a comparison between color spaces namely YUV, LAB, RGB and their effect on learned image compression. For this we use the structure and color based learned image codec (SLIC) from our prior work, which consists of two branches - one for the luminance component (Y or L) and another for chrominance components (UV or AB). However, for the RGB variant we input all 3 channels in a single branch, similar to most learned image codecs operating in RGB. The models are trained for multiple bitrate configurations in each color space. We report the findings from our experiments by evaluating them on various datasets and compare the results to state-of-the-art image codecs. The YUV model performs better than the LAB variant in terms of MS-SSIM with a Bjøntegaard delta bitrate (BD-BR) gain of 7.5\% using VTM intra-coding mode as the baseline. Whereas the LAB variant has a better performance than YUV model in terms of CIEDE2000 having a BD-BR gain of 8\%. Overall, the RGB variant of SLIC achieves the best performance with a BD-BR gain of 13.14\% in terms of MS-SSIM and a gain of 17.96\% in CIEDE2000 at the cost of a higher model complexity.
Paper Structure (14 sections, 2 equations, 6 figures, 1 table)

This paper contains 14 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related work
  3. Color spaces in learned image compression
  4. Overview
  5. Loss Function
  6. Implementation details
  7. Experiments and results
  8. Rate-distortion performance
  9. RGB image codecs
  10. YUV image codecs
  11. BD-Rate and BD-Distortion
  12. Effect of channels in chroma branch
  13. Channel impulse responses and color spaces
  14. Conclusion

Figures (6)

  • Figure 1: Network architecture of SLIC models.
  • Figure 2: RD curves of learned image codecs operating in RGB for the Kodak dataset .
  • Figure 3: RD curves of learned image codecs operating in YUV (JPEG AI and SLIC--YUV), SLIC--LAB, SLIC--RGB, and VTM for the Kodak dataset.
  • Figure 4: RD curves of chroma channel variants of SLIC--LAB and SLIC--YUV models for the Tecknick RGB dataset.
  • Figure 5: Reconstructed versions and channel impulse responses of the image GRAY_R03_0400x0400_014.png
  • ...and 1 more figures