Learning Preferential Perceptual Exposure for HDR Displays

Abstract

High dynamic range (HDR) displays are capable of displaying a wider dynamic range of values than conventional displays. As HDR content becomes more ubiquitous, the use of these displays is likely to accelerate. As HDR displays can present a wider range of values, traditional strategies for mapping HDR content to low dynamic range (LDR) displays can be replaced with either directly displaying values, or using a simple shift mapping (exposure adjustment). The latter approach is especially important when considering ambient lighting, as content viewed in a dark environment may appear substantially different to a bright one. This paper seeks to identify an exposure value which is suitable for displaying specific HDR content on an HDR display under a range of ambient lighting levels. Based on data captured with human participants, this paper establishes user preferred exposure values for a variety of maximum display brightnesses, content and ambient lighting levels. These are then used to develop two models to predict preferred exposure. The first is based on linear regression using straightforward image statistics which require minimal computation and memory to be computed, making this method suitable to be directly used in display hardware. The second is a model based on convolutional neural networks (CNN) to learn image features which best predict exposure values. The CNN model generates better results than the first model at the cost of memory and computation time.INDEX TERMS High dynamic range, machine learning, perception.

Study of naturalness in tone-mapped images

“…X is the number of observers having the same opinion about the naturalness of an image while Y is the number of observers having the opposite opinion. According this definition, there are 5 categories: (9,0), (8,1), (7,2), (6,3) and (5,4). The naturalness label of each image is decided by the majority of the observers so it appears that the confidence of the labels in category (9,0) is the highest and the confidence in category (5,4) is the lowest.…”

“…Especially, some special models used in research can reach the highest luminance value of 10,000 cd / m 2 . Although the dynamic range of new display devices is quite high, it is still quite modest when compared to the dynamic range of real scenes and the perception range of human eyes [7]. Thus, nowadays the problem of high dynamic range images is mainly related to display devices.…”