Visual comfort of binocular and 3D displays

Visual comfort of binocular and 3D displays

Frank L. Kooi, Alexander Toet*

TNO Human Factors, Vision Group, Kampweg 5, 3769 DE Soesterberg, The Netherlands

Received 9 January 2004; accepted 8 July 2004

Available online 3 August 2004

Abstract

Imperfections in binocular image pairs can cause serious viewing discomfort. For example, in stereo vision systems eye strain is caused by

unintentional mismatches between the left and right eye images (stereo imperfections). Head-mounted displays can induce eye strain due to

optical misalignments. We have experimentally determined the level of (dis)comfort experienced by human observers viewing brief

presentations of imperfect binocular image pairs. We used a wide range of binocular image imperfections that are representative for

commonly encountered optical errors (spatial distortions: shifts, magnification, rotation, keystone), imperfect filters (photometric

asymmetries: luminance, color, contrast, crosstalk), and stereoscopic disparities. The results show that nearly all binocular image

asymmetries seriously reduce visual comfort if present in a large enough amount. From our data we estimate threshold values for the onset of

discomfort. The database collected in this study allows a more accurate prediction of visual comfort from the specification of a given

binocular viewing system. Being able to predict the level of visual discomfort from the specification of binocular viewing systems greatly

helps the design and selection process. This paper provides the basis.

q 2004 Elsevier B.V. All rights reserved.

Keywords: Visual comfort; Binocular asymmetry; Binocular distortions; Binocular crosstalk; Stereoscopic vision; 3D displays

1. Introduction

This study investigates the relative contribution to visual

discomfort of spatial imperfections in binocular image pairs

that can cause viewing discomfort. In this section we will

briefly discuss the various factors that determine viewing

comfort of binocular display systems in general, and

stereoscopic viewing systems in particular, and we will

explain the goals of the present study.

1.1. Binocular display systems

Binocular display systems are nowadays found through-

out our society: in outdoor settings, office settings and in the

entertainment industry as well. The military deploy head-

mounted displays in aircraft cockpits and various types of

vehicles. 3D displays are increasingly being used to gain a

better understanding of the 3D layout of complicated

0141-9382/$ - see front matter q 2004 Elsevier B.V. All rights reserved.

doi:10.1016/j.displa.2004.07.004

* Corresponding author. Tel.: C31-3463-56237; fax: C31-3463-53977.

E-mail addresses: [email protected] (F.L. Kooi), [email protected]

(A. Toet).

structures (e.g. molecules, seismic data, anatomical data),

virtual environments, and real environments (e.g. tele-

robotics). The focus of this study is not on the technology or

application of binocular displays, but on the phenomenon

that small amounts of left/right asymmetries can dramati-

cally affect viewing comfort. These so-called binocular

errors occur not only in stereoscopic displays but in any

display that shows a different image to the left and right

eyes. There is for example an extensive literature on

left/right alignment tolerances for see-through head-

mounted displays (for a review see Ref. [1]). Another

example is the use of a monocular head-mounted display.

The left and right eyes see different images which can cause

a dramatic viewing discomfort. The issue we address here is

therefore not the quality of the percept, but the eye strain that

results from a left/right (binocular) image misalignment.

1.2. Stereoscopic display systems

A sub-category of binocular displays are the so-called

‘stereoscopic 3D displays’ which are used to create a sense

of stereoscopic depth [2,3].

Displays 25 (2004) 99–108

www.elsevier.com/locate/displa

http://www.elsevier.com/locate/displa

F.L. Kooi, A. Toet / Displays 25 (2004) 99–108100

Stereoscopic display systems offer a number of potential

benefits. The third (depth) visual dimension can in principle

be used to encode large amounts of complex information

[4]. Binocular visual input can provide a significant

improvement in perceptual separation of details and their

background, especially in unfamiliar or complex scenes [5].

Visual noise, both static and dynamic, can be separated from

the signal (actual scene objects) when two independent

images are presented in a 3D display [6]. Using binocular

vision, the observer can easily see the correlated signals

representing the objects in the scene through the uncorre-

lated noise [6]. Moreover, detection thresholds for visual

signals in noisy backgrounds can be considerably lower

when binocular cues are available [7–10]. This effect is

termed ‘binocular unmasking’. The potential advantages of

this effect have been demonstrated by using stereoscopic

video for underwater inspection under turbid conditions, a

task which is affected also by low illumination and

camouflaging [11]. Binocular unmasking can also provide

critical signal-to-noise improvement with low-light-level

intensifies imagery, forward-looking infrared, or when

looking through broken foliage. Furthermore, a stereoscopic

display can compensate for poor image quality since the

retinal disparity cues are less dependent on image quality

[12]. It can also provide a wider field of view, while

maintaining the horizontal resolution, by partially overlap-

ping the left and right-eye image [1].

However, there are also serious drawbacks associated

with the use of stereoscopic display systems. Human depth

perception is based on a combination of many visual cues

as well as internal mental templates and expectations. If

some of the 3D cues in a stereo presentation are

inconsistent with others, the perceptual system receives

conflicting information, and seeks to find a consistent

interpretation. In cases of severe conflict, 3D perception of

the scene may be totally disrupted or highly inaccurate.

Even if the user is able to perceive a consistent 3D view,

the effort required to resolve conflicts may lead to serious

fatigue, eyestrain, and headache and may reduce the sense

of realism [13–19].

In spite of the fact that a wide range of stereoscopic

display systems is currently available, and in spite of the fact

that they offer a number of potential benefits, their

application is still rather limited. The lack of viewing

comfort is one of the two primary factors limiting their

widespread use, the other factor being the nuisance of

wearing a 3D device.

1.3. Types of binocular asymmetries

Visual comfort of display systems can be seriously

reduced by many factors including jitter, flickering, image

motion, and poor resolution. In this study we exclusively

focus on visual discomfort that is caused by a difference

between the left and right images of a stereo pair; i.e. to a

binocular asymmetry. Three groups of binocular

asymmetries may be encountered in practice, depending

on the hardware of the viewing system. If optics are used the

geometry of the left and right images may differ by a shift,

rotation, magnification or by reduced resolution (image

blur). If filters are used the photometry of the left and right

images may differ in their luminance, color, sharpness

(blur), contrast, accommodation (due to chromatic aberra-

tion), or crosstalk (the left-eye view leaks through to the

right-eye view and vice versa). Thirdly, if a stereoscopic

image is presented, the stereoscopic disparity may cause an

accommodation-convergence mismatch and a motion

parallax-convergence mismatch. These mismatches do

not occur when the depth planes are optically located at

the correct viewing distance [20,21] but this is rarely the

case [6].

Another source of viewing discomfort associated with

stereo display systems is crosstalk. In a perfect stereo-

scopic display, the right-eye image should be seen by the

right-eye only and should be completely invisible to the

left-eye, and vice versa for the left-eye image. However,

complete separation of the left and right-eye images is

often impossible in 3D display systems. A percentage of

the left-eye image will be visible in the right-eye view and

vice versa. Interocular crosstalk is typically in the range of

0.1–0.3% with polarization techniques and in the range of

4 to O10% with time-multiplexed 3D displays [22]. Cross

talk produces double contours (ghosting) and is a potential

cause of eye strain and headaches [21,23]. Pastoor [24]

established visibility thresholds for cross talk using a

system with perfect left/right image separation (a high

resolution mirror stereoscope; the experimental conditions

were generated with image processing techniques). He

found that visibility of cross talk increases (i.e. the

threshold value is lowered) with increasing contrast and

increasing binocular parallax (depth) of the stereoscopic

image. In order to reproduce a reasonable depth range (up

to 40 minarc) on a high-contrast display (100:1), crosstalk

should be as low as 0.3%.

1.4. Aim of the present study

Studies that address binocular discomfort typically

address particular display devices such as HMD’s [1,25]

and stereoscopic displays [20,21]. As a result, these studies

have separately addressed the impact of respectively spatial

distortions [26], stereoscopic disparity [27], luminance

asymmetry [28], and other factors [25,29] on visual comfort.

To our knowledge no study has attempted to directly

compare the impact of the various binocular factors that

affect visual comfort. The present study aims to provide

such an overview. Knowledge of the relative importance of

these factors enables both developers and users to anticipate

and optimise the level of visual comfort associated with

binocular viewing systems in general and 3D stereoscopic

displays in particular.

F.L. Kooi, A. Toet / Displays 25 (2004) 99–108 101

2. Methods

The primary goal of this study is to assess the relative

contribution to visual discomfort of each type of binocular

asymmetry that may occur in practice. Thereto, for each

type of binocular asymmetry except for an optically induced

accommodation difference, we measured the effect on

visual comfort. In order to keep the total measuring time in

hand, the images were presented only briefly. Only mild

amounts of asymmetry were used to prevent the transfer of

eye strain from one condition to the next (context effect). To

reduce this risk even further, the order of presentation was

counter-balanced among the subjects. To obtain an estimate

of training and context effects, all subjects performed the

complete experiment twice.

2.1. Stimuli

In this study we used images that represent a part of a

typical office scene with two dominant depth planes (see

Fig. 1). The overall scene contains both highly structured and

uniform areas. The foreground of the scene displays a plant

and a book standing upright on a desktop. The background

shows partly covered bookshelves, a letter chart and a current

outlet. We recorded the scene with a digital camera. The

office scene was registered from a viewpoint at a distance of

approximately 220 cm from the office wall (the reference

position) to produce the reference scene. To produce a stereo

image pair the camera was shifted parallel to the office wall

over a distance of 3 cm to the left and 3 cm to the right

relative to the reference position. To produce two hyper-

stereo image pairs the camera was shifted over, respectively

Fig. 1. The subject views the screen (170!128 cm2) from 185 cm distance throu

projected image. With this setup, nearly all binocular errors can be simulated.

6 and 12 cm relative to the reference position. The bookcase

is 29 cm in front of the wall, and the plant is up to 55 cm in

front of the wall. This corresponds to a stereoscopic depth of,

respectively 14 and 31 arcmin, for a 6 cm camera separation.

In the 2! and 4! hyperstereo images the disparity is

increased twofold and fourfold, respectively.

During an experimental session the scene was presented

54 times to the observer, each presentation with a different

binocular asymmetry. Twenty-four subjects participated in

the experiments.

2.2. Experimental setup

A Pentium III-500 MHz computer was used to present

the stimuli and collect the observer responses. The left and

right-eye images were projected on a Miracle 3D- projection

screen by two ASK A6 XGA full color LCD projectors with

a refresh rate of 60 fields per second. The images were

displayed at a resolution of 1024!768 (horizontal by

vertical). The projection screen had an effective area of

170!128 cm2. Corresponding left and right-eye images

were projected simultaneously onto the projection screen.

Both projectors were equipped with linear polarization

filters. The filters were oriented such that the polarization of

both images was orthogonal. The subjects viewed the

images through two polarization filters: one in front of each

eye. The polarization directions of these filters were

matched to those of the image projections, such that each

eye viewed the corresponding image from the stereo pair.

Subjects were seated at a distance of 185 cm from

the projection screen. At this distance the screen subtended

a visual angle of 408.

gh two polarisation filters such that the left and right eyes see a separately

Table 1

The rating scores collected in the experiment (1, viewing comfort not

reduced; 5, extremely reduced)

No. Image manipulation LQ Median UQ

Rotations (18)

1 Out, symmetric 1.0 1.3 1.5

2 In, one eye 1.0 1.3 1.7

3 In, including border, one eye 1.0 1.4 2.0

4 Out, one eye 1.0 1.4 1.9

5 In, symmetric 1.0 1.5 2.0

Other distortions

6 1.5% Overall magnification 1.1 1.3 1.8

7 1PD trapezoid 1.3 1.5 2.0

8 3% Meridional horizontal 1.4 1.9 2.3

9 2.5% Overall magnification 1.8 2.1 2.6


The reference, normal stereoscopic and hyperstereo-

scopic images were also presented on a Silicon graphics

1600!1200 LCD monitor equipped with a Philips 9 view

autostereoscopic lenticular screen. The advantage of this

type of display is the fact that an observer does not have to

wear any optics in order to obtain a stereo percept, since the

technique to separate both views is integrated in the display.

Lenticular screens inherently produce a large amount of

crosstalk [25]. Subjects that experience reduced viewing

comfort due to crosstalk on the projection screen should

therefore also consistently give lower quality ratings to the

autostereoscopic screen.

All experiments were performed in a dimly lit room.

10 3% Meridional vertical 1.8 2.4 3.0

Shifts

11 2PD horizontal (converging) 1.5 1.6 2.6

12 1PD vertical 2.1 2.8 3.3

13 3PD horizontal (converging) 2.0 2.8 3.5

14 2PD vertical 4.0 4.5 4.9

Stereo images

15 Normal stereo (1PDZ6 cm) 1.2 2.0 2.5

16 2! Hyperstereo (1PDZ12 cm) 1.5 1.9 2.6

17 4! Hyperstereo (1PDZ24 cm) 2.0 2.5 3.8

Crosstalk

18 5% and 1PD horizontal shift 1.5 2.0 2.7



Combined manipulations

21 Stereo and 5% crosstalk 1.0 1.0 1.5

22 5% Crosstalk and blur 2.0 2.5 3.0

23 Stereo and 1PD vertical shift 2.2 2.8 3.5

Filter

24 Higher contrast (C25%) 1.0 1.3 1.4

25 Overall luminance difference 1.1 1.5 2.0

2.3. Binocular image manipulations

The original image was transformed in Adobe Photoshop

to create a set of image pairs corresponding to a range of

well defined binocular asymmetries (i.e. well specified

differences between the left and right eye images). The

transformations that were applied include rotations, scaling

and deformation operations. Some of the binocular

asymmetries involve a combination of two different types

of these image transformations. Table 1 gives an overview

of the 35 different image transformations employed in this

study. In the following we briefly describe these transform-

ations in the order of their occurrence. The transformations

were the following (the numbers in brackets refer to the

corresponding transformations in Table 1):

A 18 image rotation was created in one of several,

slightly different, ways:
(K25%)
26 Black and white representation 1.0 1.5 3.0
–
1

of t2

1=28 excyclorotation1 of the two images (1),
27 Gaussian blur (sigmaZ1 pixel) 1.7 2.2 2.7
28 Lowered contrast (K50%) 2.0 2.5 3.1
–
29 Quantization difference (4 bit) 2.0 2.8 3.3

18 incyclorotation of one image (i.e. the image to one

eye) (2),

30 Gaussian blur (sigmaZ1.5 2.3 3.0 3.5
– 18 incyclorotation of one image with border, (3)
pixels)
– 18 excyclorotation of one image (4), 31 Color asymmetry (red/green) 3.2 4.0 5.0 –
Reference image

32 Reference image 1.0 1.0 1.0

1=28 incycloratation of the two images (5).

The other distortions employed were:
Autostereoscopic display
33 No stereo 1.0 1.3 1.8
– 1.5 and 2.5% magnification of one image (6 and 9), 34 Normal stereo 2.0 3.0 3.8
35 Hyperstereo 3.4 3.8 4.5
– 3% meridional magnification (Zastigmatic effect) in the
horizontal and vertical directions (8 and 10), and
LQ, lower quartile, indicating that 25% of the subjects had a lower score. – UQ, upper quartile indicating that 25% of the subjects had a higher score.
The conditions are described in the text. 1PD, 1 prismatic dioptre which

corresponds to the angle of 1 cm viewed from 1 m distance and which

equals 0.578.

a trapezoidal distortion which vertically shrinks one side

of the image by 1PD2 (0.578) (7).

A binocular asymmetric shift was introduced as follows:

–
horizontal by 2PD and 3PD (11 and 13), and
–
vertical by 1PD and 2PD (12 and 14).
Excyclorotation is defined as a templeward turn of the upward extention

he vertical meridian (causing the top of the eye to turn out).

1PDZ1 prismatic diopterZ0.578 visual angle.

Three levels of binocular disparity were presented

(leading to stereo vision) corresponding to 6, 12, and

24 cm camera separation, or equivalently, to normal, 2!and 4! hyperstereo (conditions 15, 16 and 17 in Table 1).

For the 6 cm camera separation the resulting scene

contained 14 arcmin disparity and 31 arcmin at one location

(the plant).


Crosstalk results in a transparent overlay of the left eye

image over the right eye image or vice versa, and is

perceived as ghost, shadow or double contours. Even small

amounts of crosstalk can lead to headaches [21]. We

deployed three levels of crosstalk, corresponding to 5, 15

and 25% (conditions 18, 19 and 20 in Table 1). Since

crosstalk can only be perceived when the left and right

images are different (i.e. when the contours in both images

do not completely overlap), a small (1PD) horizontal shift

was added. To further investigate the effects of crosstalk the

reference image (condition 33 in Table 1) and two stereo-

disparity levels (conditions 34 and 35 in Table 1) were also

presented on the autostereoscopic display.

To investigate the interaction with other types of

binocular differences, we combined the following three

image manipulations:

1.

3

vie

cha

thro

sur

normal stereo and crosstalk (21),

2.
blur and crosstalk (22), and
3.

Table 2

Five point category scale used to rate the experienced viewing comfort

associated with the different binocular imperfections

a vertical shift and stereo (23).

Filters are often used to separate the left and right images

of a stereo pair for display. The use of filters can cause a

difference in contrast (conditions 24 and 28 in Table 1),

luminance (condition 25 in Table 1), sharpness, color,

and/or local luminance. We simulate the latter three

image changes by introducing blur (conditions 27 and 30

in Table 1), red/green color separation (condition 31 in

Table 1), and by reducing the color depth of one image to 4

bits, respectively (condition 29 in Table 1). The reduction of

color depth may lead to a visual impression of glossiness or

sheen, which is usually associated with metallic surfaces,

and which is sometimes called metallic luster3 [30]. As far

as we know there are no reports on the reduction of visual

comfort due to luster.

For comparison we also included the condition in which

both images of the stereo pair are represented in grayscale

only (black and white image representation: condition 26 in

Table 1).

In the reference condition (condition 32 in Table 1), the

left and right images are identical and should therefore

receive a good score.

2.4. Subjects

A total of 24 subjects, 13 males and 11 females, aged

between 17 and 58 years, served in the experiments reported

below. All subjects had normal or corrected to normal vision,

reported to have no known color deficiencies, and had a

minimum stereopsis of 60 arcsec. The subjects were

recruited within the institute (TNO Human Factors) and

Luster refers to (1) the appearance of two different surface colors

wed haploscopically and superimposed, the resulting percept being

racteristically unstable and aptly described as one surface being seen

ugh the other, and (2) a glossiness or sheen associated with metallic

faces, sometimes called metallic luster.

from a student population. They read and signed an informed

consent, and were free to leave the study at any stage.

2.5. Procedure

After signing the informed consent document, the subjects

received detailed instructions about the task they had to

perform and the attribute (eye strain) they were asked to rate.

The instructions were reiterated to ensure that the observers

understood the task at hand. The subject was instructed to

remain still during the performance of the task and to

remember the rating scale. Prior to the actual experiment the

subject performed a number of practice trials.

Each stimulus presentation consisted of three temporal

intervals:

1.

1

2

3

4

5

a 3.5 s presentation of the original (unmanipulated)

reference image, followed by

2.
a brief interval, followed by
3.
a 5 s presentation of a manipulated stereo image pair.
After viewing a stimulus presentation the subject was

asked to rate the experienced viewing comfort for the image

presented in the last temporal interval relative to the

reference image presented in the first temporal interval,

according to a five point category rating scale, adapted from

the ITU-R impairment scale [31]. This scale ranges from 1

to 5, were 1 represents no perceptible reduction in viewing

comfort compared with the reference stereo image, and 5

represents an extreme reduction in viewing comfort relative

to the reference stereo image (see Table 2).

Each subject viewed all stimulus conditions. The order

of the presentation was randomized and counter-balanced

across subjects, to reduce the possible effects of stimulus

familiarity. Randomization of the presentation order also

served to eliminate the context effect and the aftereffect of

the presentation. A high risk of subjective studies is that

previous presentations will influence the judgements over

the next presentations. The risk involves the possibility that

subjects build up a maximal amount of discomfort (for

example headaches) so that they are no longer able to make

adequate judgements. To reduce this risk, the images were

only briefly presented and the extent of the image

manipulations was kept moderate. A large number of

subjects (24) was used to reduce the abovementioned

effects as well.

A Latin square procedure was used to control the order in

which the different image manipulations were presented to

Equal viewing comfort

Slightly reduced viewing comfort

Reduced viewing comfort

Considerably reduced viewing comfort

Extremely reduced viewing comfort


the observers [32]. All the different image manipulations

were presented in a symmetric way, i.e. the same condition

was presented once to the left and once to the right eye. All

the conditions were presented in two blocks A and B: block

A contained rotations, shifts, crosstalk, stereo, mixed

combinations and filters, and block B contained the same

transformations but with opposite sign.

Subjects also had to compare images displayed on the

autostereoscopic LCD display with the reference image

displayed on the projection screen.

The overall design of the experiment was as follows:

FIRST RUN

1.
presentation of block A stimuli
2.
comparison of stimuli displayed on LCD screen to
reference stimulus on projection screen

3.
presentation of block B stimuli
4.
short break
5.
assessment of optometric status
SECOND RUN

6.
presentation of block B stimuli in counterbalanced order
7.
comparison of stimuli displayed on LCD screen to
reference stimulus on projection screen, in counter-

balanced order

8.
presentation of block A stimuli in counterbalanced order
The mean duration of an entire session of the experiment

was about 1 h.

2.6. Optometric tests

Prior to the actual experiments the optometric status of

each subject was recorded including

–
stereopsis, using the TNO test at 40 cm [33],
–
visual acuity, using a TNO Landolt C chart at a distance
of 5 m,

–
horizontal and vertical phorias at far and at near [34],
–
eye dominance (via binocular viewing through a pin-
hole), and

–
phoria, using a Maddox rod and screen at a distance of
6 m, and a Maddox rod with a self-made miniature

version of the Maddox screen at 1 m [35].

4 UQ, upper quartile indicating that 25% of the subjects had a higher

score.

3. Results

This study determines the relative importance of the

various binocular aspects that can affect the visual comfort.

The results of the experiments are listed in Table 1. Data are

presented for 35 different conditions.

No significant difference was found between the mean

rating scores resulting from the first and second run.

This implies that the experiment was neither influenced

by a strong context or learning effect. The consistency

between the two repetitions (the first and second run), which

is a good measure of the accuracy, is on average 0.5 points.

Table 1 shows the mean rating score over the 24 subjects

and the two repetitions. The quartiles give a good

impression of the inter-subject variability.

The results listed in Table 1 show that spatial distortions

that mainly affect the edges of the images of a stereo pair

turn out to cause very little discomfort. These distortions

include all (1-degree) rotations, the trapezoid distortion, and

the 1.5% magnification. This finding indicates that subjects

tend to fixate the center of the screen. We expect that these

distortions will cause more discomfort when the edges of

the screen are fixated.

The results listed in Table 1 that correspond to shifts and

meridional magnifications show that the human visual

system is clearly more sensitive for stereo image pair

mismatches in the vertical dimension than for mismatches

in the horizontal direction.

Pastoor [21] proposes the rule of thumb that a 35 arcmin

horizontal disparity is quite acceptable and 70 arcmin

disparity is too much to be viewed comfortably. This rule

agrees well with our findings for viewing stereo images (see

Table 1). The normal stereo image receives a good median

score (2.0: corresponding to a slightly reduced viewing

comfort) and an acceptable upper quartile4 score (2.5). The

4! hyperstereo image receives an acceptable median score

(2.5) and a poor upper quartile score (3.8: corresponding to a

considerably reduced viewing comfort). A significant part

of the subjects are therefore troubled by the disparities

encountered in the hyperstereo image. Notice that our test

scene primarily contains objects at two depth planes

separated by 14 arcmin disparity (the wall and the book-

cases). The only significant structure in the scene that is at

a different depth plane is the plant in the lower right, which

is 31 arcmin in front of the wall. However, the subjects

tended to fixate the central part of the scene, and mostly

ignored the plant in the lower right corner. Therefore, we

may safely assume that the characteristic disparity of this

scene in the 4! hyperstereo condition was about 4!14Z56 arcmin. This is in the middle of the transition region

between 35 and 70 arcmin as found by Pastoor [21].

Both for the polarized screen and the autostereoscopic

display, a fourfold increase in stereoscopic disparity results

in significantly degraded rating scores. However, subjects

tended to complain more about the hyperstereo condition

when viewing the autostereoscopic display. This indicates

that the crosstalk and the luster combination present in the

autostereoscopic display is equally important in causing eye

strain as the hyperstereo. These results suggest that lenticular

screens should only be used to display stereo images with

Fig. 2. Median rating scores with upper and lower quartiles for the three

levels of crosstalk. The visual comfort with 5% crosstalk is rated as ‘a bit

reduced’ (2), 15% crosstalk as ‘reduced a lot’ (3–4), and 25% as ‘extremely

reduced’ (close to 5).

Table 3

Correlations between three measures of eye quality (stereopsis, maximum

and minimum visus) and the viewing discomfort of a subset of the

conditions (vertical)

Stereopsis Visus

max

Visus min

4! Hyperstereo K0.39 (P!0.063) 0.03 0.24

Crosstalk K0.01 (P!0.95) 0.17 K0.04

Hor shift K0.33 (P!0.11) 0.19 0.23

Vert shift K0.68 (P!0.0003) 0.15 0.59 (P!0.0025)

Filters 0.11 (P!0.62) 0.17 K0.05

Autost. display K0.20 (P!0.35) 0.11 0.04

Note that the correlations with stereopsis are negative because a high value

indicates a poor score. The only binocular asymmetries that correlate

significantly with the optometric status are the vertical misalignment and

the 4! hyperstereo (we calculated the correlations based on the logarithm

of the stereopsis). Significant correlations are shown in italic. As expected,

the visual acuity of the better eye does not correlate with the rating scores.

Fig. 3. Median rating scores with upper and lower quartiles for four

binocular asymmetric shifts. H(V)1 (2,3)Zhorizontal (vertical) convergent

shift of 1 (2,3) prism diopters. Even small vertical shifts seriously degrade

the visual comfort, but a horizontal shift also makes an impact.


small amounts of disparity. Luster and crosstalk will

otherwise significantly degrade viewing comfort.

An important feature of the binocular visual system is its

ability to effectively suppress blur from the defocused eye

(interocular blur suppression: [36]). We therefore expected

that the addition of image blur in the combined conditions

would improve the rating scores. However, the results in

Table 1 show that this is not the case. This suggests that the

blur-suppression mechanism needs time to become effective.

The better rating of the ‘stereo-&-crosstalk’ condition

than of the ‘1PD-shift-&-crosstalk’ condition confirms that

the discomfort due to crosstalk scales with the amount of

stereoscopic disparity [27].

Of the filter manipulations, color separation, blur, and

luster (resulting from the reduction of the color depth of one

image to 4-bits) have the most impact on viewing comfort.

The viewing comfort associated with the reference image

is rated as ‘not degraded’, just as expected.

To provide more insight into the data, a part of the results

is also shown graphically in Figs. 2 and 3. Fig. 2 shows

the median rating scores with upper and lower quartiles for

the three levels of crosstalk. The viewing comfort of images

containing 5% crosstalk is rated as ‘a bit reduced’ (2), for

15% crosstalk it is considered as ‘reduced a lot’ (3–4), and

for 25% crosstalk it is experienced as ‘extremely reduced’

(close to 5). The three conditions differ statistically

significant from each other (P!0.05). Fig. 3 shows the

median rating scores with upper and lower quartiles for four

different horizontal and vertical binocular asymmetric

shifts. This figure clearly shows that the human visual

system is most sensitive (in terms of comfort) to vertical

binocular asymmetries. Even a small vertical shift seriously

degrades the visual comfort. However, horizontal shifts also

degrade stereoscopic viewing comfort.

3.1. Optometry: between-subject differences

We were particularly interested whether the level of

experienced visual discomfort is related to the optometric

status of the viewer. One can imagine, for example, that a

subject with the tendency to squint could be bothered more

by a spatial distortion. We therefore correlated the

optometric data (visual acuity, stereopsis, phoria) with the

rating scores. The results are listed in shown in Table 3.

Subjects with good vision are bothered more by an image

misalignment than the subjects with poor vision. The reason

is easy to understand: a person with limited binocular vision

is less able to see the misalignment. We do not find an effect

of the subject’s eye quality on the image manipulations that

are also noticeable with one eye. These include crosstalk

and the filter conditions. The virtual absence of a correlation

with the visual acuity of the better eye (VAmax) indicates

that the quality of binocular vision is determined by

the poorer eye. This is confirmed by the high correlation

between a subject’s stereopsis score and the visual

acuity of the poorer eye (VAmin) on the one hand and the

absence of a correlation between stereopsis and VA max on

the other hand. These correlations are K0.96 and C0.3,

respectively.

Table 4

Threshold values for each of the binocular image manipulations

Image manipulation Threshold value

Rotations (18) O18

Other distortions

Overall magnification 2.5%

Trapezoid O1PD

Meridional horizontal O3%

Meridional vertical !3%

Shifts

Horizontal Between 2 and 3PD

Vertical !1PD

Stereo images Between 2! and 4! hyper-stereo

Crosstalk

Low disparity O5%

High disparity 5%

Filter

Contrast difference Between 25 and 50%

Luminance difference (–25%) O25%

Black and white representation OK

Gaussian blur !1 pixel

Local luminance Between 4 and 8 bit digitization

Color asymmetry Less than red/green separation

The thresholds are derived from the data in Table 1. As threshold we take a

median score of 2 (viewing comfort is ‘a bit reduced’ for 50% of the

observers) or an upper quartile score of 3 (viewing comfort is ‘reduced’ for

25% of the observers). The conditions are described in the text.


In summary, the quality of a person’s binocular vision

affects the binocular viewing comfort to a limited extent.

Binocular misalignment and excessive stereoscopic dis-

parity are less troublesome for people with reduced

(binocular) vision.

3.2. Binocular image manipulations thresholds

The data in Table 1 can be used to estimate thresholds for

each of the binocular image manipulations. As threshold we

choose to adopt a median score of 2 (viewing comfort is ‘a

bit reduced’ for 50% of the observers) or an upper quartile

score of 3 (viewing comfort is ‘reduced’ for 25% of the

observers). The resulting thresholds are listed in Table 4.

For some of the image manipulations it is not possible to

give accurate threshold estimates from our data. In that case

we list the highest value of the respective distortion for

which the observers experienced no reduction in viewing

comfort, and we indicate that the threshold is larger than the

listed value.

For example, Table 4 shows that the observers

experienced no reduction in viewing comfort for rotations

up to 18. This implies that the actual threshold is larger than

18, and designers should not be worried when image

rotations up to 18 occur in practice.

From Table 4 we see that, in order to prevent reduced

viewing comfort with differentially distorted binocular

image pairs,

–
an overall image magnification should be less than 2.5%,
–
the meridional vertical distortion should be less than 3%,
–
a horizontal shift should be less than about 2 and 3PD,
–
a vertical shift should be less than 1PD,
–
the disparity in a stereoscopic image should not exceed
30 armin,

–
for high disparity images the crosstalk should be less than
5%,

–
the contrast difference should not exceed 25%,
–
the Gaussian blur should have a standard deviation less
than 2.3 armin (corresponding to 1 pixel on our display),

–
color asymmetry should be less severe than complete
red/green separation,

–
for 8 bit imagery the quantization difference should be
less than 4 bits,

and that viewing comfort will not be reduced for

–
a trapezoidal image distortion up to 1PD
–
a meridional horizontal distortion up to 3%,
–
low disparity images if the crosstalk is up to 5%,
–
global luminance differences up to 25%,
–
a black and white image representation.
4. Discussion and conclusions

The present study addresses static images that are

presented for short amounts of time. This condition is

representative for many actual viewing situations since

people tend to regularly alternate fixation between a display

and other parts of the visual world. A notable exception is

formed by immersive HMD’s that prevent this behaviour by

occluding everything except the display itself. Although

display duration has a clear effect on stereoscopic vision in a

performance-oriented context [37–39], it has been shown

before that brief image presentations suffice to obtain

reliable appreciative judgements of static stereoscopic

image material [12,13].

The factors that determine stereoscopic viewing comfort

most strongly are vertical disparity, crosstalk, and blur.

Even a little amount of each (1PD, 5%, 3.4 arcmin,

respectively) causes noticeable discomfort. Hyperstereopsis

may cause some discomfort when the represented depth

exceeds the depth-of-focus as previously reported [27], but

the effect is rather weak. For short viewing durations, the

distortions that only affect the periphery of the perceived

image are less important.

The quality of a person’s binocular vision affects the

binocular viewing comfort to a limited extent. Binocular

misalignment and excessive stereoscopic disparity is less

troublesome for people with reduced (binocular) vision.

The database collected in this study allows a more

accurate prediction of visual comfort from the specifica-

tion of a given binocular viewing system. Being able

to predict the level of visual discomfort from the

specification of binocular viewing systems greatly helps


the design and selection process. This paper provides the

basis.

In conclusion, our results indicate that 3D displays

should be produced with sufficient accuracy to eliminate all

binocular image errors that cause viewing discomfort.

Lenticular screens should only be used for the display of

stereoscopic images with relatively small disparities, since

this type of 3D display inherently produces crosstalk and

luster.

References

[1] M.E. Kalich, C.E. Rash, C. van de Pol, T.L. Rowe, L.M. Lont,

R.D. Peterson, Biocular image misalignment tolerance in: C.E. Rash,

C.E. Reese (Eds.), Helmet- and Head-Mounted Displays VIII:

Technologies and Applications, The International Society for Optical

Engineering, Bellingham, WA, 2003, pp. 284–295.

[2] S. Pastoor, M.D. Wopking, 3-D displays: a review of current

technologies, Displays 17 (2) (1997) 100–110.

[3] I. Sexton, Ph. Surman, Stereoscopic and autostereoscopic display

systems—an in-depth review of past, present and future technologies,

IEEE Signal Processing Magazine 16 (3) (1999) 85–99.

[4] J. Liu, S. Pastoor, K. Seifert, J. Hurtienne, Three-dimensional PC:

toward novel forms of human-computer interaction in:

S.P.K.S.J.H. Liu (Ed.), Three-Dimensional Video and Display:

Devices and Systems SPIE CR76, Boston, MA USA (2000).

[5] W.R. Watkins, G.D. Heath, M.D. Phillips, J.M. Valeton, A. Toet,

Search and target acquisition: single line of sight versus wide baseline

stereo, Optical Engineering 40 (9) (2001) 1914–1927.

[6] F.L. Kooi, A. Toet, Additive and subtractive transparent depth

displays in: J.G. Verly (Ed.), Enhanced and Synthetic Vision, The

International Society for Optical Engineering, Bellingham, WA,

USA, 2003, pp. 58–65.

[7] B. Schneider, G. Moraglia, A. Jepson, Binocular unmasking: an

analog to binaural unmasking?, Science 243 (4897) (1989) 1479–

1481.

[8] G. Moraglia, B. Schneider, On binocular unmasking of signals in

noise: further tests of the summation hypothesis, Vision Research 32

(2) (1992) 375–385.

[9] G. Moraglia, B. Schneider, Binocular unmasking with vertical

disparity, Canadian Journal of Psychology 45 (3) (1991) 353–366.

[10] G. Moraglia, B. Schneider, Effects of direction and magnitude of

horizontal disparities on binocular unmasking, Perception 19 (5)

(1990) 581–593.

[11] S. Ma, P. Milgram, Investigation of potential benefits of stereoscopic

video for visual detection in turbid underwater environments in:

A.J. Woods, M.T. Bolas, J.O. Merritt, S.A. Benton (Eds.), Stereo-

scopic Displays and Virtual Reality Systems XIII, SPIE—The


2001, pp. 22–33.

[12] W.A. IJsselsteijn, H. de Ridder, J. Vliegen, Subjective evaluation of

stereoscopic images: effects of camera parameters and display

duration, IEEE Transactions on Circuits and Systems for Video

Technology 10 (2) (2000) 225–233.

[13] W.A. IJsselsteijn, H. de Ridder, J. Vliegen, Effects of stereoscopic

filming parameters and display duration on the subjective assessment

of eye strain in: J.O. Merritt, S.A. Benton, A.J. Woods, M.T. Bolas

(Eds.), Stereoscopic Displays and Virtual Reality Systems VII, The


2000, pp. 12–22.

[14] J.D. Pfautz, Sampling artifacts in perspective and stereo displays in:

A.J. Woods, M.T. Bolas, J.O. Merritt, S.A. Benton (Eds.),

Stereoscopic Displays and Virtual Reality Systems XIII, SPIE—The


2001, pp. 54–62.

[15] M. Siegel, Y. Tobinaga, T. Akiya, Kinder gentler stereo, in:

J.O. Merritt, M.T. Bolas, S.S. Fisher (Eds.), Stereoscopic Displays

and Virtual Reality Systems VI, The International Society for Optical

Engineering, Bellingham, WA, 1999, pp. 18–27.

[16] M. Siegel, Just enough reality: a kinder gentler approach to stereo in:

D.G. Hopper (Ed.), Cockpit Displays VI: Displays for Defense

Applications, The International Society for Optical Engineering,

Bellingham, WA, 1999, pp. 173–179.

[17] M. Siegel, Perceptions of crosstalk and the possibility of a zoneless

autostereoscopic display in: A.J. Woods, M.T. Bolas, J.O. Merritt,

S.A. Benton (Eds.), Stereoscopic Displays and Virtual Reality

Systems XIII, SPIE—The International Society for Optical Engin-

eering, Bellingham, WA, 2001, pp. 34–41.

[18] L. Meesters, W.A. IJsselsteijn, P. Seuntiens, A survey of perceptual

quality issues in three-dimensional television systems in: A.J. Woods,

M.T. Bolas, J.O. Merritt, S.A. Benton (Eds.), Stereoscopic Displays

and Virtual Reality Systems X, The International Society for Optical

Engineering, Bellingham, WA, USA, 2003, pp. 313–326.

[19] L.B. Stelmach, W.J. Tam, D.V. Meegan, Perceptual basis of

stereoscopic video in: J.O. Merritt, M.T. Bolas, S.S. Fisher (Eds.),

Stereoscopic Displays and Virtual Reality Systems X, The Inter-

national Society for Optical Engineering, Bellingham, WA, USA,

1999, pp. 260–265.

[20] S. Pastoor, Human factors of 3D displays, Displays 1993; 14.

[21] S. Pastoor, Human factors of 3D imaging: results of recent research at

Heinrich-Hertz-Institut Berlin Proceedings of the International Dis-

play Workshop ’95 (Asia Display ’95) (1995).

[22] T. Honda, Dynamic holographic 3D display using LCD Asia Display

95 (1995), 1995, pp. 777–780.

[23] Y.Y. Yeh, L.D. Silverstein, Limits of fusion and depth judgement in

stereoscopic color displays, Human Factors 32 (1) (1990) 45–60.

[24] S. Pastoor, Human factors of 3D imaging: results of recent research at

Heinrich-Hertz-Institut Berlin Proceedings of the International Dis-

play Workshop ’95 (Asia Display ’95) (1995).

[25] M. Velger, Helmet-mounted displays and sights Artech House,

Norwood, MA, 1998.

[26] H.C. Self, Optical tolerances for alignment and image differences for

binocular helmet-mounted displays Armstrong Aerospace Medical

Research Laboratory, Wright-Patterson AFB, OH, 1986 (Report

AAMRL-TR-86-019).

[27] S. Pastoor, Human factors of 3D displays in advanced image

communications, Displays 14 (3) (1993) 150–157.

[28] I.P. Beldie, B. Kost, Luminance asymmetry in stereo TV images in:

J.O. Merritt, S.S. Fisher (Eds.), Stereoscopic Displays and Appli-

cations II, SPIE—The International Society for Optical Engineering,

Bellingham, WA, 2003, pp. 242–247.

[29] C.E. Rash, Helmet mounted displays: design issues for rotary-wing

aircraft U.S. Army Aeromedical Research Laboratory (USAARL), Fort

Rucker, AL, 2003 (Report available on http://www.usaarl.army.mil/).

[30] D. Cline, H. Hofstetter, J.R. Griffing, Dictionary of visual science and

related clinical terms Butterworth-Heinemann, Oxford, UK, 1980.

[31] ITU-R, Methodology for the subjective assessment of the quality of

television pictures International Telecommunication Union, Geneva,

Swiss, 2003 (Report BT.500-9).

[32] W.A. Wagenaar, Note on the construction of digram-balanced Latin

squares, Psychological Bulletin 72 (6) (1969) 384–386.

[33] J. Walraven, TNO Test for Stereoscopic Vision Lameris Instrumen-

ten, Utrecht, The Netherlands, 1972.

[34] W. Duke-Elder, Text-Book of Ophthalmology Vol. IV Henry Kimton,

London, UK, 1949.

[35] J.B. Eskridge, J.F. Amos, J.D. Bartlett, Clinical Procedures in

Optometry J.B. Lippincott Company, Philadelphia, USA, 1991.

[36] C.M. Schor, P. Erickson, Ocular dominance accommodation and

the interocular suppression of blur in monovision, in: G. Obrecht,

http://www.usaarl.army.mil/


L. Stark (Eds.), Presbyopia Research: from Molecular Biology

to Visual Adaption, Plenum Press, New York, USA, 1991,

pp. 273–288.

[37] W.J. Tam, L.B. Stelmach, Display duration and stereocopic depth

discrimination, Canadian Journal of Experimental Psychology 52

(1998) 56–61.

[38] R. Patterson, R. Cayko, L. Short, R. Flanagan, L. Moe, E. Taylor,

P. Day, Temporal integration differences between crossed and

uncrossed stereoscopic mechanisms, Perception and Psychophysics

57 (1995) 891–897.

[39] R. Patterson, R. Fox, The effect of testing method on stereoanomaly,

Vision Research 24 (1984) 403–408.

Visual comfort of binocular and 3D displays

Documents

Transcript of Visual comfort of binocular and 3D displays