Zoned Mapping Network from SDR Video to HDR Video

HDR video is popular for its brilliant colors, incredible brightness and rich details. Considering the strict production conditions of HDR video, most videos are currently saved in SDR format. Converting massive SDR videos into HDR format through technical methods can quickly fill the vacancy of HDR videos in the market and improve the utilization efficiency of film and television resources. In this paper, we propose a zoned mapping network to convert the video format. According to SDR OETF, the video frames are firstly segmented into three regions, highlight, medium and darkness. Then a tone mapping model is designed and respectively trained for the above three regions. Moreover, a detail enhancement model is proposed to form a composite inverse tone mapping network. Experiments show that our method can not only achieve good subjective visual quality, but also accomplish excellent results on various objective metrics, such as PSNR, SSIM, VIF, etc.

I. Introduction

Iteration of camera sensor and display backlight techniques has enabled people to record and reproduce more realistic scene with High Dynamic Range (HDR) signal. For a scene with rich brightness, which human eyes are most sensitive to, HDR video preserves more details.