A Generative Adversarial Network Based Tone Mapping Operator for 4K HDR Images

Authors:

Abstract
High dynamic range (HDR) has arguably been established as the preferred image and video format for content providers. As standard dynamic range (SDR) displays still dominate the market, there is a need for finding efficient ways to convert HDR content to the SDR format, a process known as tone mapping. Recently, many tone mapping operators (TMOs) have been proposed that are based on deep learning approaches. However, the biggest challenge in training such deep learning networks is lack of truthful SDR and HDR datasets that would lead to highly accurate TMOs. In this paper, we introduce a new high-quality 4K HDR-SDR dataset of image pairs, covering a wide range of brightness levels and colors. We propose a TMO that is based on the generative adversarial network architecture. Evaluation results showed that our method achieves high perceptual quality, maintaining artistic intent and providing better color representation compared to existing state-of-the-art TMOs. Data and code are available at: https://github.com/zjbthomas/TMO-GAN.
Funding Information
  • Natural Sciences and Engineering Research Council of Canada

This publication has 12 references indexed in Scilit: