Since the ratification of ITU-R Recommendation BT.2100 in 2017 [10], the production and distribution of high dynamic range (HDR) content has increased exponentially. Industry creatives have leveraged a new set of tools that expose audiences to increased detail in both deeply dark and expansively bright imagery. Creatives may want to intentionally exploit the extended luminance range and introduce high-magnitude changes within their content. While this represents a positive advancement for artistic expression, it also increases the potential for sudden, unintended luminance jumps when programs are edited together. There are several cases in which abrupt, large changes in luminance are neither authored by creatives nor preferred by audiences. Examples include changes that occur at program transitions, advertisement insertions, social media mixed feeds, and channel changes. These unintended changes in luminance both disrupt the creative intent of the original content and provide the audience with an unsatisfactory experience. It is of interest to understand the underlying phenomena driving the audience’s reaction in these scenarios.

Quantifying the magnitude of the luminance jump provides the first indication of how impactful the transition is going to be. However, simply recognizing the differences in luminance level at either end of the transition cannot, alone, characterize the visual experience. The direction of the change will also impact the nature of the reaction. A transition that increases in luminance will provide a different experience than one that decreases in it. This article highlights the adaptation time in which the audience was subjected to one luminance level before switching to another. Adaptation time is defined as a third, major contributing factor to the audience’s response. New subjective testing is described to evaluate the effects of increasing and decreasing adaptation time on the comfort level experienced during HDR luminance transitions.

There is an abundance of research describing luminance perception and adaptation. Of particular interest to this study is the human visual system’s reaction to the sequential presentation of various images with distinguishably different luminance levels. This phenomenon involves understanding the interpretation of mean display luminance levels and exploring the influence of adaptation.

As discussed, the subjective experiments were designed to evaluate the effects of the luminance-change magnitude as well as the adaptation time. Also considered was the direction of the change (darker or brighter). Understanding that dark adaptation is slower than light adaptation, there is an expectation for transitions reflecting a dark-to-bright change to result in a different observer response than those reflecting a bright-to-dark one. Because of the impact the direction of the luminance change has on the subject’s adaptation, the experiment was split into two parts. The first part tested the dark-to-bright scenario, while the second part tested the bright-to-dark one. Splitting the experiment into two parts both ensured a consistent adaptation procedure and prevented observer fatigue. It is highly possible that the results would have skewed, had the observers been asked to rapidly switch between a predominantly dark adaptation state and a predominantly bright one between trials.

When designing the workflow, it was desired to keep the adaptation procedure consistent, trial to trial. Instead of presenting two images (one to adapt to and one to change to) per trial, we presented three images. The first and third images were the same. By having the primary adaptation happen on the second of the three images, this workflow prevented the adaptation state of the subjects from varying. This design also allowed for the experiment to closely mimic a channel change operation. The luminance transitions simulated a situation where an observer starts on one channel, switches to another one, and finally changes back to the starting one.

Figure 1 illustrates the experimental procedure that was followed for the bright-to-dark-to-bright adaptation version of the experiment. After providing demographic information and issuing consent to participate in the test, each subject was brought into a 5 cd/m2 surround environment. Before the main experiment, the subjects were given instructions and participated in a short practice round. Each trial started by showing a bright image for ten seconds. Following this, a dark image was presented for a variable amount of time. Finally, the bright image was shown again for seven seconds, and the observer was asked to rate the visual experience. The ratings were based off the observers’ immediate sensation when the dark image switched back to the bright one. The dark-to-bright-to-dark adaptation version of the experiment followed a similar procedure. First, a dark image was presented for ten seconds, followed by a bright image for a variable amount of time, before the dark image returned to the screen for seven seconds. The observers, once again, rated the sensation immediately following the switch from the bright image back to the dark one.

To test the influence of adaptation time, the duration in which the observer is subjected to the second channel varied between trials. As discussed, dark and light adaptation are processed at different speeds. For this reason, it was deemed necessary for the two versions of the experiment to test different adaptation durations. A series of beta studies determined the optimal durations for each version; shown in Table I.

As previously mentioned, subjects were asked to rate their experience following the second transition. To gauge the impact of the transition, observers were asked to make their selections based on their immediate reaction to the transition. Originally, the five-grade impairment scale described in ITU-R BT.500 [11] was used. During the beta testing period, however, observers struggled to apply this rating scale to the experimental task.

Because all images exhibited notably different mean display luminance values, observers were confused with the score of “5,” which denotes “imperceptible difference.” The observers also felt limited in their responses for the ratings that indicate an “annoying” experience. Many of our subjects asked to have more gradation in those choices. In addition to feedback about the mechanics of the impairment scale, observers felt the term “annoying” did not apply to the bright-to-dark-to-bright transitions. It was better understood when the experience was described as “painful,” to match the physical sensation in their eyes. However, the observers accepted the term “annoying” for the dark-to-bright-to-dark transitions. The observers felt it appropriately described the inability to discern dark detail following bright adaptation.

Using a combination of the feedback from beta observers and influence from the eleven-grade numerical scale specified in ITU-R BT.500, a new seven-grade impairment scale was implemented. The scores and terminology affiliated with the new impairment scale are shown in Table II. There are two values given to terms that indicate a degree of pain or annoyance to add gradation in the responses. As mentioned, the two versions of the test use different language to reflect the observer feelings following the luminance transition.

A total of seventeen participants completed the experiment. Ten of the participants were female, while the other seven were male. Fifteen of the participants were in either their twenties or thirties, while the other two were in their fifties. Ten of the participants were familiar with either color science or image evaluation. Seven observers had not previously participated in a subjective test. All seventeen of the participants were informed of the experiment’s procedure before both versions and agreed to take part in it twice.

Within the two versions of the experiment, three different transitions were evaluated between bright and dark images across different adaptation time periods. By modelling the results, we explore the influence adaptation time brings to the mean subjective score (MSS) for the two experiment conditions. The fitted trends were approximated by implementing a model derived by Smirnakos et al. [20]. The linear luminance values were scaled to PQ and time was sampled based on image frames. The time constants were calculated based on a frame rate of 24 fps. where t is the frame number and τ is the characteristic time of decay; a constant which is determined from:

Figures 4 and 5 show the MSS values and 95% confidence intervals for the three images across the nine adaptation time trials. Fig. 4 corresponds to the bright-to-dark-to-bright experiment, while Fig. 5 reflects the dark-to-bright-to-dark one. As mentioned previously, the ratings scale was created based on impairment scales presented in ITU-R BT.500. An MSS of 4 represents a transition which does not introduce any disruption in observer comfort, while scores at or below 2 suggest a noteworthy change in experience.

A repeated-measures analysis of variance (ANOVA) test was run to compare the mean selections for the three luminance-change cases in each of the two experiments. Statistical significance was confirmed in the results of all three luminance levels within a 95% confidence interval [18] across two degrees of freedom for both experiments. The results of the bright-to-dark-to-bright experiment exhibited an F-statistic of 8.1 and a p-value of 0.002, while those of the dark-to-bright-to-dark experiment exhibited an F-statistic of 83.6 and a p-value of 1.004 × 10−10.

The results of the bright-to-dark-to-bright experiment confirm that the MSS values are influenced by the magnitude of the luminance change and the adaptation time. The data suggests that observers will find the experience more comfortable when the magnitude of the difference is lowered and/or the adaptation time is shortened. The MSS values begin to reach a plateau for all three cases at the largest adaptation time values. As mentioned previously, this was expected based on the initial testing done to select the adaptation time durations and prior modelling of this phenomenon.

The results of the dark-to-bright-to-dark experiment similarly confirm that the MSS values take influence from the luminance change magnitude and the adaptation time. All three cases in this experiment, however, exhibit some level of annoyance at the one-half second duration. Even at this short duration, the observers experienced perceptual afterimage artifacts caused by the bright image. At this one-half second trial, the annoyance level was seen to be higher for each of the three cases. This confirms the influence of luminance-change magnitude on observer responses and suggests a non-linear relationship. Despite the different starting points, the results of all three cases ease into a plateau at the higher time course values.

Looking, first, at the case where the observers switch from the 20 cd/m2 image to the dark image in Fig. 5, the results convey a comfortable experience for all trials. Only in the last trial, which subjected observers to a one-minute adaptation time course, there was an indication of slight annoyance by the MSS values. Where the observers adapt to the 100 cd/m2 image and the 338 cd/m2 image, a decrease in the comfort level was immediately seen. This trend of increasing the luminance-change magnitude was expected. Being adapted to a brighter luminance level will increase the time necessary to shift viewer adaptation down to a darker one. Following the decrease in luminance, the observers found the experience of waiting to see the details in the dark image unfavorable. The larger the magnitude of the luminance change, the more annoying the experience. Adapting to the 100 cd/m2 image saw slight annoyance in the MSS values, while adapting to the 338 cd/m2 image saw MSS values indicating notable annoyance. Despite following a non-linear trend, the observer results align with the findings of Noland and Pindoria at the 10 second mark (matching their experiment design).

While this experiment shined light on how observers respond to sudden luminance changes, it does not cover the problem in full. As discussed, the observer response to luminance changes was evaluated along the lines of the changes’ magnitude, direction, and adaptation time. Aligning these components with one another in different respects allowed for their individual influences to take impact on comfort level. One major difference between the two versions of the experiment was the descriptive language of the MSS choices. While swapping out “annoyance” with “pain” helped observers distinguish the difference between the cases, it may have made the results between the two versions disproportionate. It’s entirely possible that an observer didn’t experience any eye pain during a bright-to-dark-to-bright trial but was somewhat annoyed by the change. This level of scrutiny does not exist in the presented data. It would be helpful to ensure that the responses between the two versions are distributed evenly.

In the context of this study, participants were instructed to respond to the immediate change in luminance. While this provides insight into the reaction following the transition, it does not cover the duration of the initial response or how it changes over time. It is likely that the level of discomfort becomes lower very quickly after the bright-to-dark-to-bright transition. Similarly, few observers might not mind the dark-to-bright-to-dark transition, if the details in the dark image become visible shortly after the change. This phenomenon, in a sense, represents a secondary adaptation parameter that exists following the luminance change. Further investigation into the longevity of the initial response can help complete the visual model.

As stated previously, the study was designed to mimic a channel change operation. While the emphasis was on the second transition that brought the observer back to the previous “channel,” no consideration was given to either the first transition or how the two transitions operated together. Considering the experience collectively could cater the response to an evaluation of why the second channel “interrupts” the first one. For example, if bright content interrupted dark content for a very short duration, it would likely be viewed as more visually annoying than if it lasted a long duration. It’s also possible that the opposite results would occur in the reverse situation of dark content unexpectedly interrupting bright content. Expanding the problem’s scope from one transition in luminance to a more-complete scenario can provide insight into additional components of the visual experience.

This research explored three luminance-transition components that influence observer responses: the magnitude of the of change, the direction of the change (darker or brighter), and the adaptation time before the change. The presented results confirm that all three components provide significant influence. Increasing or decreasing the magnitude of the luminance change will impact the comfort level of the observer in a similar way to increasing or decreasing the adaptation time. In television and social media, there are often program transitions, advertisement insertions, and channel change operations. We have shown that these unexpected transitions have the potential to cause visual discomfort or annoyance. A surprising result was the non-linear subjective responses with the dark-to-bright-to-dark transition, which fell sharply with little adaptation time. Further study is needed to verify what caused this. Additional opportunities for further study include investigating the longevity of the response after the luminance change and extending the problem to include multiple luminance transitions that occur sequentially. Furthermore, the influences of color palette and local adaptation are yet to be explored. Regardless, this research provided insight into how human subjects respond to unintended luminance changes across varying periods of adaptation.