Deep Learning-Based HDR Image Upscaling Approach for 8K UHD Applications

Advances in display technology have led to the introduction of 8K Ultra High Definition (UHD) displays to the consumer market, offering an improved visual experience. However, the lack of 8K High Dynamic Range (HDR) content is a major challenge for the wide adoption. In this paper, we introduce a deep learning approach based on generative adversarial networks to generate 8K UHD HDR content from Full High Definition and 4K content. Benefiting from a multiple-level residual and dense structure, along with a random down-sampling method, our approach yields natural and visually pleasing 8K UHD HDR content with consistent color performance.

I. Introduction

Rapid advances in display technologies have led to the re-cent introduction of 8K Ultra High Definition (UHD) high dynamic range (HDR) displays to the consumer market. These HDR-enabled 8K UHD TVs are promising cinematic viewing experience to the end user, offering ultra-high resolution, extended BT.2020 color gamut, and increased peak luminance values [1]. However, while 8K UHD content creation is still a costly and time-consuming process, most manufactures are turning to upscaling existing Full High Definition (FHD) and 4K UHD content in order to enable this new market. To this end, many methods based on recent emergence of deep learning (DL) that focus on upscaling standard dynamic range (SDR) images have been developed in the past few years [2]–[3].