High dynamic range (HDR) image rendering has been an active research area in the last two decades. In real-world scenarios, there is a wide range of luminance (around 14 log units) between highlights and shadows. However, according to previous reports, the simultaneous dynamic range of human vision is 3.73 log cd/m. This property suggests that it is unnecessary to compress the full dynamic range of luminance in a real-world scene for realistic rendering. In this study, we present a sequence of psychophysical experiments to determine an appropriate simultaneous dynamic range depending on the HDR scene using an HDR rendering algorithm. The experimental results are summarized as follows: (1) for an HDR scene within the range of human vision, a rendered image with a higher simultaneous dynamic range within this range is preferable, (2) for an HDR scene beyond the range of human vision, a rendered image with the same simultaneous dynamic range as that of human vision is preferable, and (3) a rendered image with a higher simultaneous dynamic range is preferably observed under ambient light rather than in a dark room. INTRODUCTION Owing to the popularization of the personal computer and the Internet in recent years, we have many opportunities to view digital images in daily life. We now use digital cameras as the most common imaging devices. However, a display monitor as a commonly used image display device has a narrow range of luminance (at most, only about 3 log units), while the real world has a much wider range of luminance (around 14 log units). Therefore, it is difficult to reproduce an HDR scene on the display. The technique that compresses the dynamic range to display an HDR scene on an LDR monitor is called “tone mapping,” and various methods have been proposed so far. In many of these methods, the S-shaped function [1] modeled by Naka and Rushton has been used. This model uses a sensitivity parameter that can change the simultaneous dynamic range that the human can perceive. However, the parameter had to be empirically determined. In this study, we performed subjective evaluation experiments with HDR scenes to clarify the relationship between the dynamic range of the actual HDR scene and the simultaneous dynamic range for realistic HDR rendering [2] using the S-shaped function. RENDERING ALGORITHM In this study, we used the HDR rendering algorithm proposed in Ref.[2] to reproduce the images used in the evaluation experiments. The algorithm is a spatially variant operator for imitating the Spotential function and realizing the local adaptation process as follows:
[1]
K. Naka,et al.
S‐potentials from luminosity units in the retina of fish (Cyprinidae)
,
1966,
The Journal of physiology.
[2]
Takahiko Horiuchi,et al.
HDR Image Quality Enhancement Based on Spatially Variant Retinal Response
,
2010,
EURASIP J. Image Video Process..
[3]
J. Rabin.
The Retina: An Approachable Part of the Brain
,
2013
.
[4]
Erik Reinhard,et al.
A reassessment of the simultaneous dynamic range of the human visual system
,
2010,
APGV '10.