Tone Mapping HDR Panoramas for Viewing in Head Mounted Displays

Miguel Melo

, Kadi Bouatouch

, Maximino Bessa

1,3

, Hugo Coelho

, Remi Cozot

and Alan Chalmers

INESC TEC, Porto, Portugal

IRISA, University of Rennes I, Rennes, France

UTAD, Vila Real, Portugal

WMG, University of Warwick, Coventry, U.K.

Keywords:

Head Mounted Displays, HDR Panoramas, Tone Mapping, High Dynamic Range.

Abstract:

Head-mounted displays enable a user to view a complete environment as if he/she was there; providing an

immersive experience. However, the lighting in a full environment can vary signiﬁcantly. Panoramic images

captured with conventional, Low Dynamic Range (LDR), imaging of scenes with a large range of lighting

conditions, can include areas of under- or over-exposed pixels. High Dynamic Range (HDR) imaging, on

the other hand, is able to capture the full range of detail in a scene. However, HMDs are not currently HDR

and thus the HDR panorama needs to be tone mapped before it can be displayed on the LDR HMD. While a

large number of tone mapping operators have been proposed in the last 25 years, these were not designed for

panoramic images, or for use with HMDs. This paper undertakes a two part subjective study to investigate

which of the current, state-of-the-art tone mappers is most suitable for use with HMDs.

1 INTRODUCTION

Panoramic images capture the full 360

◦

of detail in

a scene (Figure 1). When shown within a Head-

Mounted Display (HMD) the viewer can look around

the captured environment as if he/she was there, pro-

viding a highly immersive experience. There can be

a wide range of lighting in any scene, from the sun to

dark shadows. Traditional panoramic imaging tech-

niques, known as Low Dynamic Range (LDR) are

only able to capture a dynamic range of up to 256 to

1 (8 stops) and thus will fail to capture all the detail,

resulting in areas of the panoramic image with under-

or over-exposed pixels. By using 32 bit IEEE ﬂoating

point values to represent each colour channel, High

Dynamic Range (HDR) imaging is able to capture the

full range of lighting in a scene.

Worldwide sales of Head Mounted Displays

(HMDs) are increasing rapidly with cumulative to-

tal sales expected to top 200 million from 2015 to

2020 (Tractica, 2017) and the revenue for VR re-

lated hardware is due to exceed $3.6 billion in 2017

(Nafarrete, 2017). The low cost availability of high

quality HMDs, such as the Oculus Rift, HTC Vive,

etc., means that these devices are now being used for

a wide range of applications from computer games

to ﬁlms. In particular 360

◦

content is becoming

more widely available helped by companies, such as

YouTube which provides the ability to upload and

◦

videos on their website in March 2015.

Currently HMDs are not HDR, and thus an HDR

image has to be tone mapped before it is displayed

within the HMD. A large number of tone mapping

operators (TMOs) have been presented in the last 25

years (Banterle et al., 2011). However, these TMOs

were not designed for panoramic images and how the

perception of a scene may change as the user looks

around within the environment. This paper under-

takes a detailed two part subjective evaluation of the

use of TMOs for 360

◦

HDR panoramic images. In the

ﬁrst study, participants were asked to rank according

to their preference a set of tone mapped HDR panora-

mas without reference. In the second part of the

study, the same participants were asked to rank non-

panoramic images tone mapped with the same opera-

tors within the HMD against the reference HDR im-

age displayed on an HDR display. As can been seen in

Figure 1, a 360

◦

HDR panoramic image displayed on

an HDR display would be distorted and thus not be a

fair comparison with the image as it would be viewed

in the HMD. In addition, the study also compared the

performance of two makes of HMD, the Oculus Rift

232

Melo, M., Bouatouch, K., Bessa, M., Coelho, H., Cozot, R. and Chalmers, A.

Tone Mapping HDR Panoramas for Viewing in Head Mounted Displays.

DOI: 10.5220/0006615402320239

In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 1: GRAPP, pages

232-239

ISBN: 978-989-758-287-5

Figure 1: HDR panorama.

DK2 and the Oculus Rift CV1, to investigate whether

the technical speciﬁcations of the HMD itself had an

inﬂuence on the choice of TMO for the HMD.

2 RELATED WORK

Tone mapping attempts to preserve the perception of

a scene, captured with an HDR image, on an LDR

display. Many TMOs have been proposed since the

ﬁrst one was published in 1993 (Chiu et al., 1993),

For a detailed overview of these see (Banterle et al.,

2011). TMOs are typically classifed as global or lo-

cal. Global TMOs apply the same operation to all

pixels in the image. They are typically able to run

in real time. Local TMOs, on the other hand, ap-

ply apply the operator differently to each pixel, taking

into account the pixels around the pixel being consid-

ered. Local TMOs tend to be much more complex

than global ones but they are, unlike global TMOs,

able to preserve both global and local contrast. An

additional category temporal was recently introduced

to take into account how a tone mapper may need to

deal with how a scene changes over time, for example

when tone mapping HDR video (Banterle et al., 2011;

Boitard et al., 2014).

In this paper four state-of-the-art TMOs, which

have previously been shown to be very effective on a

number of different displays (Melo et al., 2014), were

used:

Fer: Developed by Ferweda et al. (Ferwerda et al.,

1996), this TMO includes a model of visual adap-

tation which takes into account aspects of the Hu-

man Visual System (HVS), including visibility,

visual acuity, and illumination changes adapta-

tion. Threshold-versus-intensity (TVI) functions

are used to model photopic and scotopic vision,

while a linear combination of both photopic and

scotopic vision is used for mesopic vision.

Man: This display adaptive TMO uses a model of

the HVS to minimise visible contrast distortions

(Mantiuk et al., 2008). In addition, this TMO

takes into account the ambient light levels and dis-

play characteristics.

Boi: Designed especially for HDR video, this is a

temporal coherent TMO that preserves overall

contrast (Boitard et al., 2012). This is achieved

by considering the perceptual consistency of an

object throughout the video. A frame is processed

in two steps. In the ﬁrst step the frame is tone

mapped with (any) global tone mapper, while in

the second step, the luminance of the frame is ad-

justed taking into account the lighting of the entire

video.

Pat: Known as the time-dependent visual adaptation

TMO (Pattanaik et al., 2000), this method takes

into account that the HVS does not react instantly

to large changes in luminance levels. The TMO

simulates how the HVS responds to real world

lighting levels.

2.1 Evaluating Tone Mappers

The different TMOs have been evaluated in detail

over the years using both objective and subjective

metrics. The most widely used objective metric has

been VDP (Daly, 1993) and its HDR versions, HDR-

VDP (Mantiuk et al., 2004) and HDR-VDP-2 (Man-

tiuk et al., 2011). Other objective metrics include

VQM (Narwaria et al., 2015) and the dynamic range

independent quality assessment metric (Aydin et al.,

2010).

One of the ﬁrst subjective evaluations of TMOs

was by Drago et al. (Drago et al., 2002) in which

Tone Mapping HDR Panoramas for Viewing in Head Mounted Displays

233

4 HDR scenes were tone mapped with 7 different

TMOs. Ledda et al. (Ledda et al., 2005) were the ﬁrst

to use an HDR display as a reference. Tone mapped

images displayed on two LDR displays on either side

of the HDR display were compared with the HDR im-

age on the HDR display. Other evaluations of TMOs

include work by (Yoshida et al., 2005), (Kuang et al.,

2007), (

Cad

ık et al., 2008) and (Melo et al., 2014;

Melo et al., 2015). Recently temporal TMOs have

been subjectively evaluated by Eilertsen et al. (Eilert-

sen et al., 2013).

2.2 Tone Mapping for HMDs

Cutchin et al. presented an interactive view depen-

dent tone mapping technique for HDR panoramas on

HMDs via a view-adjusted mapping function stored

in a separate texture ﬁle, which they termed Tone-

Texture (Cutchin and Li, 2016). This technique was

used to expand the perceived colours within the HMD

in order to improve the panorama’s visual appear-

ance. The approach was shown to provide a bet-

ter visual appearance compared to the panorama tone

mapped with the Photographic TMO (Reinhard et al.,

2002). They provided demonstration systems for We-

bGL and the Oculus Rift and GearVR HMDs.

Yu presented a dynamic TMO for HDR panora-

mas which takes into account the user’s head position

and thus the viewport (Yu, 2015). A simple model

of eye adaptation is used to simulate light and dark

adaptation. Yu’s TMO was shown in a study of only

8 participants to be preferred over the global Photo-

graphic tone mapper (Reinhard et al., 2002).

Perrin et al. (Perrin et al., 2017) undertook a qual-

ity assessment of 360

◦

HDR content when displayed

in HMDs. Three global (Simple linear scaling, Pho-

tographic (Reinhard et al., 2002), Display adaptive

(Mantiuk et al., 2008))) and two local TMOs (Detail

preserving (Mantiuk et al., 2006) and Exposure fusing

(Mertens et al., 2009)) were evaluated using a simple

pairwise comparison. There results showed that the

participants did not have a clear preference for any of

the TMOs evaluated.

3 SUBJECTIVE EVALUATION

This subjective study evaluated four state-of-the-art

TMOs across two consumer-grade HMDs, namely

Oculus RIFT CV1 and Oculus RIFT DK2. The

study was divided into two. The ﬁrst experiment

consisted of evaluating ﬁve HDR panorama images

viewed on the HMD. These panoramas had each been

tone mapped with the four TMOs. Participants were

asked to record their preference. The second experi-

ment asked the participants to rank the TMOs applied

to set of HDR images viewed on an HMD, compared

to the HDR images displayed on a SIM2 HDR dis-

play. A participant only performed the experiments

on one of the two HMDs (between subjects) but ev-

ery participant performed the rankings for both HDR

panoramas and HDR images (within-subjects).

3.1 Sample

A total of 15 participants with ages between 20 and

44 (M = 27.62, SD = 7.38 ) were divided into two

groups. One group performed the experiments using

the Oculus RIFT CV1 (N=7) and the other group per-

formed the experiments using the Oculus RIFT DK2

(N=8). All participants reported normal or corrected-

to-normal vision.

3.2 Variables

There were three independent variables: the HMDs,

the Content Type, and the selected TMOs. The depen-

dent variable is the TMO ranking provided by partic-

ipants. Ranking was preferred over rating as it avoids

narrow distributions by guarantying that each ranked

item has a unique value for all tone mapped stimuli.

Ranking was also chosen over pair-wise comparisons

as this would require a signiﬁcantly larger number of

comparisons to compare all stimuli.

3.3 Materials

The HMDs used for the experiments were the Oculus

Rift CV 1 that has a resolution of 2160 × 1200, a re-

fresh rate of 90Hz and a ﬁeld of view of 110

◦

and the

Oculus Rift DK2 with a resolution of 980 × 1080, a

refresh rate of 75Hz and a ﬁeld of view of 100

◦

. The

stimuli consisted of 5 HDR panoramas, and 5 HDR

images

. These were tone-mapped using the 4 dif-

ferent TMOs: Fer, Man, Boi and Pat, as described in

section 2.

Custom VR software was developed to enable

participants to perform the experiments with ease.

This software consisted of different screens (one per

each panorama) where participants could watch ev-

ery tone-mapped panorama the number of times they

needed to perform the rankings. For ranking the tone

mapped panoramas, participants had to drag the cor-

responding tone mapped panorama thumbnails to a

set of ranking boxes numbered from 1 to 4 where 1

Please refer to http://www.miguelmelo.pt/hdrhmd/ for

the images used in this study.

GRAPP 2018 - International Conference on Computer Graphics Theory and Applications

234

was the preferred (or closer to the reference in ex-

periment 2) and 4 the less preferred (or less close to

the reference in experiment 2). The experiments were

conducted using a desktop PC with an Intel i7-5820K

CPU, a NVIDIA GeForce GTX 980 Ti graphics card

and 16GB of RAM.

3.4 Procedure

The experiments were conducted in a laboratory set-

ting where all the environment variables such as

sound, temperature and lighting levels were con-

trolled. The ﬁrst step was to present the experimen-

tal study to the participants without disclosing its pur-

pose to avoid bias. A consent form was given to par-

ticipants for them to agree to take part in the study.

How the software worked was then explained and a

demonstration scene instructed them how to perform

the two-part experiments.

• Part 1: Participants were asked to rank 4 TMOs

applied to the HDR panoramas according to their

preference taking into consideration overall image

quality, naturalness and detail across 5 scenes. In

this experiment, there was no reference.

• Part 2: Participants were asked to rank 4 TMOs

applied to HDR images according to a reference

that was shown on an HDR display across 5

scenes.

The experimental team helped participants take the

HMDs on and off and remained in the experimental

room in case participants had any questions and to

ensure that the data was collected properly. Partici-

pants were free to drop-out from the experiments at

any given time. On average, the entire procedure took

approximately 15 minutes.

3.5 Statistical Procedures

To study the impact of HMDs across Content Types

over TMOs rankings, a 2 (HMD) × 2 (Content Type)

× 4 (TMOs) mixed design factorial ANOVA was per-

formed where TMO was a within-participant vari-

able and HMD and Content Type were between-

participant variables. A Boxs M test was used to

test the multivariate homogeneity of the data. To in-

vestigate the performance of TMOs across the dif-

ferent scenarios, a Mann-Whitney U was performed.

The Friedman test was carried out to verify if TMOs

rankings were statistically signiﬁcantly different in

each scenario. Additionally, post hoc analysis with

Wilcoxon signed-rank tests was conducted with a

Bonferroni correction applied, to identify possible

groupings of TMOs. Kendalls Coefﬁcient of Con-

cordance W was also used to give an estimate of

the agreement amongst participants. A Coefﬁcient

of Concordance of W = 1 signiﬁes perfect agreement

amongst the participants and W = 0 indicates com-

plete disagreement.

4 RESULTS

For ease of understanding, the results are divided into

subsections: Results across all HMDs across all con-

tent types; HMD impact on TMO performance for

HDR panoramas; HMD impact on TMO performance

for HDR images; Content type impact on TMOs per-

formance on Oculus RIFT CV1; and ,Content type

impact on TMOs performance on Oculus RIFT DK2.

4.1 Results Across All HMDs Across All

Content Types

To test the multivariate homogeneity, a Box’s M test

was applied which revealed that the within-group co-

variance matrices are equal and that a MANOVA

can be applied (Box’s M = 1.511, p > 0.05). The

MANOVA analysis showed that a statistically signiﬁ-

cant impact of the HMDs over TMOs ranking can be

assumed but with some reservation as there is only a

moderate effect F(4,143)= 2.369, p = 0.055, Wilks

= .938, η

= .062, O.P. =.673. For Content Type, sta-

tistically signiﬁcant differences were found in TMOs

rankings, F(4,143)= 5.668, p = 0.000, Wilks

λ =

.859, η

= .141, O.P. = .981. For HMD × Con-

tent Type, no statistically signiﬁcant differences were

found, F(4,143)= .994, p = 0.413, Wilks

λ = .973, η

= .027, O.P. = .308.

4.2 HMD Impact on TMO Performance

for HDR Panoramas

A Mann-Whitney U test showed that there was no sta-

tistically signiﬁcant difference in the performance of

TMOs across the two HMDs for the HDR panoramas:

• Man: U = 596.000, p = .227

• Fer: U = 641.500, p = .511

• Boi: U = 553.500, p = .105

• Pat: U = 558.500, p = .081

As there were no differences of performance across

the two HMDs, the results for the two scenarios were

grouped to perform a Friedman test to analyze if

there were differences across the TMOs. The Fried-

man test revealed a statistically signiﬁcant difference

between the different TMOs (χ

(3) = 70.041, p =

0.000). Post hoc analysis with Wilcoxon signed-rank

Tone Mapping HDR Panoramas for Viewing in Head Mounted Displays

235

Table 1: Wilcoxon Signed-Rank Test results across HMDs

for HDR panoramas.

Paired Comparison Z p

Fer-Man -4.849 0.000

Boi-Man -2.662 0.008

Pat-Man -6.338 0.000

Boi-Fer -2.152 0.031

Pat-Fer -3.979 0.000

Pat-Boi -5.320 0.000

tests was conducted with a Bonferroni correction ap-

plied, resulting in signiﬁcance levels p < .031 for all

the comparisons as shown in table 1. The TMOs rank-

ing order was: Man (M=1.73), Boi (M=2.13), Fer

(M=2.67), and Pat (M=3.41) with no groupings of

TMOs. Kendalls coefﬁcient of Concordance was W

= 0.311 (p < 0.05).

4.3 HMD Impact on TMO Performance

for HDR Images

A Mann-Whitney U test showed that there was no sta-

tistically signiﬁcant difference in the performance of

TMOs across the two HMDs for the HDR images:

• Man: U = 566.500, p = .113

• Fer: U = 624.500, p = .394

• Boi: U = 607.000, p = .295

• Pat: U = 635.500, p = .436

Again, as there were no differences of performance

across the two HMDs for HDR images, the results

for the two scenarios were grouped to perform a

Friedman test to analyze if there were differences

across the TMOs. The Friedman test revealed a sta-

tistically signiﬁcant difference between the different

TMOs (χ

(3) = 70.041, p = 0.000). Post hoc anal-

ysis with Wilcoxon signed-rank tests was conducted

with a Bonferroni correction applied, resulting in sig-

niﬁcance levels across all TMOs except for the pair

Fer - Man (p = .884) as shown in table 2. The TMOs

ranking order was: Man (M=1.93), Fer (M=1.95),

Boi (M=2.69), and Pat (M=3.44), with Man and Fer

grouped together. Kendalls coefﬁcient of Concor-

dance was W = 0.312 (p < 0.05).

4.4 Content Type Impact in TMOs

Performance on Oculus RIFT CV1

A Mann-Whitney U test showed that there was no sta-

tistically signiﬁcant difference in the performance of

TMOs across the two content types:

• Man: U = 523.500, p = .258

Table 2: Wilcoxon Signed-Rank Test results across HMDs

for HDR images.

Paired Comparison Z p

Fer-Man -0.146 0.884

Boi-Man 3.755 0.000

Pat-Man -6.512 0.000

Boi-Fer -3.314 0.001

Pat-Fer -5.906 0.000

Pat-Boi -3.701 0.000

Table 3: Wilcoxon Signed-Rank Test results across content

types for Oculus Rift CV1.

Paired Comparison Z p

Fer-Man -0.146 0.064

Boi-Man 3.755 0.016

Pat-Man -6.512 0.000

Boi-Fer -3.314 0.498

Pat-Fer -5.906 0.000

Pat-Boi -3.701 0.000

• Fer: U = 452.000, p = .051

• Boi: U = 564.000, p = .554

• Pat: U = 588.000, p = .750

As there were no differences of performance

across the two content types when using the Ocu-

lus Rift CV 1, the results for the two scenarios were

grouped to perform a Friedman test and verify if

there were differences across the TMOs. The Fried-

man test revealed a statistically signiﬁcant difference

between the different TMOs (χ

((3) = 35, 584, p =

0.000). Post hoc analysis with Wilcoxon signed-

rank tests was conducted with a Bonferroni correc-

tion applied, resulting in signiﬁcance levels across all

TMOs except for the pair Fer - Man (p = .064) and

Boi Fer (p = 495) as shown in table 3. The TMOs

ranking order was: Man (M=1.99), Fer (M=2.31),

Boi (M=2.45), and Pat (M=3.24), with Man and Fer

grouped together as well as Fer and Boi. Kendalls co-

efﬁcient of Concordance was W = 0.169 (p < 0.05).

4.5 Content Type Impact on TMOs

Performance on Oculus RIFT DK2

The Mann-Whitney U test conducted to identify if

there were differences of TMO performance across

the two types of content have revealed that there are

statistically signiﬁcance differences for Fer and Boi:

• Man: U = 691,000, p = .241

• Fer: U = 386,000, p = .000

• Boi: U = 471,000, p = .001

GRAPP 2018 - International Conference on Computer Graphics Theory and Applications

236

Table 4: Wilcoxon Signed-Rank Test results for HDR

panoramas for Oculus Rift DK2.

Paired Comparison Z p

Fer-Man -4.560 0.000

Boi-Man -1.810 0.070

Pat-Man -5.463 0.000

Boi-Fer -2.292 0.022

Pat-Fer -3.771 0.000

Pat-Boi -4.825 0.000

Table 5: Wilcoxon Signed-Rank Test results for HDR im-

ages for Oculus Rift DK2.

Paired Comparison Z p

Fer-Man -4.560 0.948

Boi-Man -1.810 0.000

Pat-Man -5.463 0.000

Boi-Fer -2.292 0.001

Pat-Fer -3.771 0.000

Pat-Boi -4.825 0.007

• Pat: U = 730,500, p = .425

As there were signiﬁcant differences on TMOs perfor-

mance across the two content types, the TMOs rank-

ing was analyzed separately. For the HDR panora-

mas, a Friedman test revealed a statistically signiﬁ-

cant difference between the different TMOs (χ

(3) =

57, 316, p = 0.000). Post hoc analysis with Wilcoxon

signed-rank tests was conducted with a Bonferroni

correction applied, resulting in signiﬁcance levels

across all as shown in table 4. The TMOs rank-

ing order was: Man (M=1.60), Boi (M=2.01), Fer

(M=2.76), and Pat (M=3.63), with Man and Boi

grouped together. Kendalls coefﬁcient of Concor-

dance was W = 0.478 (p < 0.05).

Regarding HDR images, the Friedman test re-

vealed that there are statistically signiﬁcant dif-

ferences between the different TMOs (χ

(3) =

52, 010., p = 0.000). Post hoc analysis with Wilcoxon

signed-rank tests was conducted with a Bonferroni

correction applied, resulting in signiﬁcance levels

across all as shown in table 5. The TMOs rank-

ing order was: Man (M=1.78), Fer (M=1.85), Boi

(M=2.83), and Pat (M=3.55), being that Man and Fer

were grouped together. Kendalls coefﬁcient of Con-

cordance was W = 0.433 (p < 0.05).

5 DISCUSSION

This paper has investigated three important research

questions: 1) Is the choice of TMO for displaying

an HDR image affected by the use of an HMD for

the display?; 2) Does content type have an effect

on the choice of TMO for use with an HMD?; and

3)Is the TMO performance consistent across HMDs

across Content Types?

Table 6 summarizes the TMOs Rankings and

groupings across scenarios from the analysed results.

From the statistics of the results we obtained for

our user studies, the following is clear:

• The choice of HMD seems to have an impact

on TMO ranking: Although MANOVA analy-

sis revealed a p = 0.055, the null hypothesis is

rejected with some reservation as there is only a

moderate effect of the sample and an OP of near

67%. This latter result corroborates previous stud-

ies such as (IJsselsteijn et al., 2000), (IJsselsteijn

et al., 2001), (Lin et al., 2002), (Ba

nos et al.,

2008), and (Young et al., 2014) which have iden-

tiﬁed technology is a factor that inﬂuences a VR

experience. Our paper extends this previous work

to now also include the inﬂuence of technology

when displaying HDR content on HMDs.

• The Content Type has impact on TMO rank-

ing: Depending on the HDR content to be shown,

one should carefully select the appropriate TMO.

For instance, for HDR panoramas, Boi performs

better than Fer but the inverse happens when the

content is HDR images. If one considers the fea-

tures of both TMOs, the results seem to indicate

that local TMOs perform better for HDR panora-

mas while global TMOs are more suited for HDR

images. This might happen because of the dis-

tortion of the source panorama images; a local

processing would consider that distortion and per-

form better tone mapping due use of the neigh-

bouring pixels’ information.

• TMOs performance is consistent across scenar-

ios (except for Content Type when the Oculus

Rift DK2): Despite the impact of the different

HMDs and/or content types on the TMOs perfor-

mance, the same TMO will have similar perfor-

mance across conditions.

• Concordance between participants on rank-

ings are higher for Oculus Rift DK2 (CV 1 ≈

.20 vs DK2 ≈ .45): Participants were able to

identify more differences in CV1 between TMOs

therefore there was less concordance. This sug-

gests that less complex TMOs can be used for

older devices as the difference between TMOs is

not so noticeable.

From the statistics of the results we obtained for

our user studies, the global MANOVA and the set of

pairwise comparisons, the following is clear:

• The choice of HMD has impact on TMO ranking

Tone Mapping HDR Panoramas for Viewing in Head Mounted Displays

237

Table 6: Overall results obtained for each scenario. Coloured groupings represent TMOs that were not found to be signiﬁcantly

different using pairwise comparisons to each other, via Bonferroni adjustment, at p < 0.01.

Kendall’s

Co-efﬁcient

of Concordance* 1

Rankings across HMDs

for HDR Panoramas**

0.311

Man Boi Fer Pat

Rankings across HMDs

for HDR images**

0.312

Man Fer Boi Pat

Rankings across

Content Type for CV1**

0.169

Man Fer Boi Pat

Rankings for HDR panoramas

using Oculus Rift DK2

0.478

Man Boi Fer Pat

Rankings for HDR images

using Oculus Rift DK2

0.433

Man Fer Boi Pat

for (p < 0.01)

The independent variables were grouped as there was no statistically signiﬁcant difference

between conditions,p < 0.005.

• The content Type (panorama × image) has impact

on TMO ranking

• Concordance between participants on rankings

are much higher for DK2 (CV1 ∼ 20 vs DK2

∼ 45). In particular, participants where able to

identify more differences in CV1 between TMOs

therefore there was less concordance. This sug-

gests that less complex TMOs can be used for

older devices as the difference between TMOs is

not so noticeable.

6 CONCLUSIONS

With the rapid growth of VR technology and the in-

creasing demand for more realism within VR experi-

ences, more content is likey to be produced in HDR,

as this has no areas of under- or over-exposed pixels.

This is also very likely to be HDR panoramic content.

Until HMDs include HDR displays, this HDR content

will need to be tone mapped for enhanced display on

the LDR displays of the HMD. This paper has shown

that we cannot rely on previous evaluations of TMOs

to chose the right TMO for use with HDR panoramic

content for HMDs. A new choice of TMO speciﬁ-

cally for HMDs is necessary to achieve a high level of

perceived realism in the HMD, especially when con-

sidering panoramic images.

The subjective evaluation has considered only 5

HDR panoramas. To provide conclusive results more

HDR panoramas need to be considered. In addition,

these panoramas should be chosen from a mixture of

scenes, such as outdoor daylight, outdoor night light,

indoor lighting, mixture of indoor and outdoor lights,

etc. We will investigate this in the future as we wish

to investigate whether the ﬁnal preference of a TMO

may depend not only on the HMD but also on the con-

tent, for which we will need to capture this more com-

plete HDR panorama database.

Future work will also investigate, including with

the use of eyetrackers, whether more in-depth knowl-

edge of how humans view content on HMD screens,

could lead to even more effective TMOs for display-

ing HDR panoramic content on HMDs than the ones

preferred in our experiments.

ACKNOWLEDGEMENTS

Part of the work was developed at MASSIVE VR

Laboratory established by INESC TEC through the

project REC I/EEI-SII/0360/2012 entitled “MAS-

SIVE - Multimodal Acknowledgeable multiSenSory

Immersive Virtual Environments” ﬁnanced by the Eu-

ropean Union (COMPETE, QREN and FEDER).

REFERENCES

Aydin, T. O., Cadik, M., Myszkowski, K., and Seidel, H.-P.

(2010). Video quality assessment for computer graph-

ics applications. In SIGGRAPH Asia. ACM.

nos, R. M., Botella, C., Rubi

o, I., Quero, S., Garc

ıa-

Palacios, A., and Alca

niz, M. (2008). Presence and

emotions in virtual environments: The inﬂuence of

stereoscopy. CyberPsychology & Behavior, 11(1):1–

Banterle, F., Artusi, A., Debattista, K., and Chalmers, A.

(2011). Advanced High Dynamic Range imaging.

CRC Press.

Boitard, R., Bouatouch, K., Cozot, R., Thoreau, D., and

Gruson, A. (2012). Temporal coherency for video tone

mapping. SPIE Optical Engineering+ Applications,

pages 84990D–84990D.

GRAPP 2018 - International Conference on Computer Graphics Theory and Applications

238

Boitard, R., Cozot, R., Thoreau, D., and Bouatouch, K.

(2014). Zonal brightness coherency for video tone

mapping. Signal Processing: Image Communication,

29(2):229–246.

Cad

ık, M., Wimmer, M., Neumann, L., and Artusi, A.

(2008). Evaluation of HDR tone mapping methods

using essential perceptual attributes. Computers &

Graphics, 32(3):330–349.

Chiu, K., Herf, M., Shirley, P., Swamy, S., Wang, C., and

Zimmerman, K. (1993). Spatially Nonuniform Scal-

ing Functions for High Contrast Images. In Proceed-

ings of Graphics Interface ’93, pages 245–253.

Cutchin, S. and Li, Y. (2016). View dependent tone map-

ping of HDR panoramas for head mounted displays.

In Proceedings of the 26th International Conference

on Artiﬁcial Reality and Telexistence and the 21st

Eurographics Symposium on Virtual Environments,

ICAT-EGVE ’16, pages 29–36. Eurographics Associ-

ation.

Daly, S. (1993). The visible differences predictor: an al-

gorithm for the assessment of image ﬁdelity. pages

189–206. Digital Image and Human Vision.

Drago, F., Martens, W., Myszkowski, K., and Seidel,

H.-P. (2002). Perceptual evaluation of tone map-

ping operators with regard to similarity and pref-

erence. Research Report MPI-I-2002-4-002, Max-

Planck-Institut f

ur Informatik, Stuhlsatzenhausweg

85, 66123 Saarbr

ucken, Germany.

Eilertsen, G., Unger, J., Wanat, R., and Mantiuk, R. (2013).

Survey and evaluation of tone mapping operators for

HDR video. In ACM SIGGRAPH 2013 Talks.

Ferwerda, J. A., Pattanaik, S. N., Shirley, P., and Green-

berg, D. P. (1996). A model of visual adaptation for

realistic image synthesis. In Proceedings of the 23rd

annual conference on Computer graphics and inter-

active techniques, pages 249–258. ACM.

IJsselsteijn, W., de Ridder, H., Freeman, J., Avons, S. E.,

and Bouwhuis, D. (2001). Effects of stereoscopic pre-

sentation, image motion, and screen size on subjec-

tive and objective corroborative measures of presence.

Presence, 10(3):298–311.

IJsselsteijn, W. A., de Ridder, H., Freeman, J., and Avons,

S. E. (2000). Presence: concept, determinants, and

measurement. In Electronic Imaging, pages 520–529.

International Society for Optics and Photonics.

Kuang, J., Yamaguchi, H., Liu, C., Johnson, G. M., and

Fairchild, M. D. (2007). Evaluating HDR rendering

algorithms. ACM Trans. Appl. Percept., 4(2).

Ledda, P., Chalmers, A., Troscianko, T., and Seetzen, H.

(2005). Evaluation of tone mapping operators using

a high dynamic range display. In ACM SIGGRAPH

2005 Papers, SIGGRAPH ’05, pages 640–648, New

York, NY, USA. ACM.

Lin, J.-W., Duh, H. B.-L., Parker, D. E., Abi-Rached, H.,

and Furness, T. A. (2002). Effects of ﬁeld of view

on presence, enjoyment, memory, and simulator sick-

ness in a virtual environment. In Virtual Reality, 2002.

Proceedings. IEEE, pages 164–171. IEEE.

Mantiuk, R., Daly, S., and Kerofsky, L. (2008). Display

adaptive tone mapping. In ACM Transactions on

Graphics (TOG), volume 27, page 68. ACM.

Mantiuk, R., Kim, K. J., Rempel, A. G., and Heidrich, W.

(2011). HDR-VDP-2: A calibrated visual metric for

visibility and quality predictions in all luminance con-

ditions. ACM Trans. Graph., 30(4):40:1–40:14.

Mantiuk, R., Myszkowski, K., and Seidel, H.-P. (2004).

Visible difference predicator for high dynamic range

images. In Proceedings of IEEE International Confer-

ence on Systems, Man and Cybernetics, pages 2763–

2769.

Mantiuk, R., Myszkowski, K., and Seidel, H.-P. (2006). A

perceptual framework for contrast processing of high

dynamic range images. ACM Transactions on Applied

Perception, 3(3):286–308.

Melo, M., Bessa, M., Debattista, K., and Chalmers, A.

(2014). Evaluation of HDR video tone mapping for

mobile devices. Signal Processing: Image Commu-

nication, Special Issue on Advances in High Dynamic

Range Video Research, 29(2):247–256.

Melo, M., Bessa, M., Debattista, K., and Chalmers, A.

(2015). Evaluation of Tone-Mapping Operators for

HDR Video Under Different Ambient Luminance

Levels. Computer Graphics Forum, 34(8):38–49.

Mertens, T., Kautz, J., and Van Reeth, F. (2009). Exposure

fusion: A simple and practical alternative to high dy-

namic range photography. Computer Graphics Forum,

28(1):161–171.

Nafarrete, J. (2017). Report: Global VR hardware revenue

to hit $3.6 billion in 2017.

Narwaria, M., Silva, D., and Le Callet, P. (2015). HDR-

VQM: An objective quality measure for high dynamic

range video. Signal Processing: Image Communica-

tion, 35:46–60.

Pattanaik, S. N., Tumblin, J., Yee, H., and Greenberg, D. P.

(2000). Time-dependent visual adaptation for fast re-

alistic image display. In Proceedings of the 27th an-

nual conference on Computer graphics and interac-

tive techniques, pages 47–54. ACM Press/Addison-

Wesley Publishing Co.

Perrin, A.-F. N. M., Bist, C., Cozot, R., and Ebrahimi, T.

(2017). Measuring quality of omnidirectional high

dynamic range content. In Optics+Photonics Opti-

cal Engineering+Applications Applications of Digital

Image Processing XL. SPIE.

Reinhard, E., Stark, M., Shirley, P., and Ferwerda, J. (2002).

Photographic tone reproduction for digital images. In

Proceedings of SIGGRAPH 2002, pages 267–276.

Tractica (2017). More than 200 million virtual reality head-

mounted displays to be sold by 2020.

Yoshida, A., Blanz, V., Myszkowski, K., and Seidel, H.-P.

(2005). Perceptual evaluation of tone mapping oper-

ators with real-world scenes. In 17th Annual Sympo-

sium on Electronic Imaging. SPIE.

Young, M. K., Gaylor, G. B., Andrus, S. M., and Bo-

denheimer, B. (2014). A comparison of two cost-

differentiated virtual reality systems for perception

and action tasks. In Proceedings of the ACM Sym-

posium on Applied Perception, pages 83–90. ACM.

Yu, M. (2015). Dynamic tone mapping with head-mounted

displays. Standford University Report.

Tone Mapping HDR Panoramas for Viewing in Head Mounted Displays

239