Automatic detection of issues related to colour vision deficient internet users

(1)

IN

DEGREE PROJECT COMPUTER SCIENCE AND ENGINEERING, SECOND CYCLE, 30 CREDITS

,

STOCKHOLM SWEDEN 2017

Automatic detection of issues related to colour vision deficient internet users

JOEL EKMAN

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF COMPUTER SCIENCE AND COMMUNICATION

(2)

English title

Automatic detection of issues related to colour vision deficient internet users

Swedish title

Automatisk identifiering av problem relaterade till internetanvändare med defekt färgseende

Author

Joel Ekman, joelekm@kth.se

Submitted for the completion of the KTH program;

Human Computer Interaction, Master of Science in Computer Science and Engineering Supervisor: Rebekah Cupitt, KTH, School of Computer Science and Communications, Department of Media Technology and Interaction Design

Examiner: Henrik Artman, KTH, School of Computer Science and Communications, Department of Media Technology and Interaction Design

Date of submission: 2017-06-19

(3)

Abstract

With increasing demand placed on online accessibility, a tool that enables developers to more easily build accessible websites for colour vision deficient (CVD) internet users becomes a crucial consideration. An extension was developed for the web browser, Google Chrome, and issues for CVD internet users were identified. The identification was based on the luminance and colour contrast between different objects next to each other on a web page, such as fonts and their background. The extension calculated how a CVD internet user would experience the colours and then checked the contrast between them. The extension's calculations and reliability was assessed through a evaluation with CVD internet users and the results

suggested that it would be possible to reliably detect issues related to CVD internet users with a algorithm implemented in a Chrome extension.

Sammanfattning

Med ökande krav på tillgänglighet på internet så ökar behovet av ett verktyg som underlättar för utvecklare att bygga hemsidor för personer med defekt färgseende. Ett tilläggsprogram utvecklades för webbläsaren Google Chrome, och problem för personer med defekt färgseende identifierades. Identifikationen baserades på luminanskontrast och färgkontrast mellan olika närliggande objekt på en hemsida, så som fonter och deras bakgrund.

Tilläggsprogrammet beräknade hur en person med defekt färgseende upplever färgerna på hemsidan och kontrollerar kontrasten mellan dem. Tilläggsprogrammets tillförlitlighet bedömdes med hjälp av en utvärdering genomförd med personer med defekt färgseende.

Utvärderingen indikerade att det är möjligt att tillförlitligt identifiera problem relaterade till personer med defekt färgseende med hjälp av en algoritm implementerad i ett Chrome tilläggsprogram.

(4)

Automatic detection of issues related to colour vision deficient internet users

Joel Ekman

KTH Royal Institute of Technology Stockholm, Sweden

joelekm@kth.se

ABSTRACT

With increasing demand placed on online accessibility, a tool that enables developers to more easily build accessible websites for colour vision deficient (CVD) internet users becomes a crucial consideration. An extension was developed for the web browser, Google Chrome, and issues for CVD internet users were identified. The identification was based on the luminance and colour contrast between different objects next to each other on a web page, such as fonts and their background. The extension calculated how a CVD internet user would experience the colours and then checked the contrast between them. The extension’s calculations and reliability was assessed through a evaluation with CVD internet users and the results suggested that it would be possible to reliably detect issues related to CVD internet users with a algorithm implemented in a Chrome extension.

Keywords

Colour vision deficiency, Colour blindness, Colour contrast, Developer tools, Accessibility

1. INTRODUCTION

There is an increasing demand for online accessibility, as can be seen in the new European Union legislation [19]

that requires that all physical needs are considered for state projects. Colour vision deficiency or more commonly called colour blindness affects 8% of all men and 0.5% of women [23]. The lack of consideration for this large group of people is something that can become costly once the Swedish law about online accessibility is enforced. The first case is currently being tried in court[16]. The increase in demand for accessibility comes from both businesses that wish to avoid paying fines and from a society that is trying to make the internet more accessible. This makes it an interesting opportunity in developing a tool to help reach this goal.

Today, tools for finding issues related to colour vision deficient internet users are based on visualising how the CVD person might see a web page. The developer then looks at

ACM ISBN . DOI:

the web page and checks through it using a heuristic algorithm to identify the potential issues. While relatively effective, this is required for each of the eight types and severities of colour vision deficiencies and is time demand- ing. Based on the researcher’s experience working with web development this is not always done.

1.1 Colour Vision Deficiency

Simunovic [23] gives a good introduction to the subject.

He writes that a human without colour vision deficiencies (CVD) has vision that is trichromatic, that is any colour is built up by a mixture of the three primary colours red, green, and blue. These are registered by three different cone photoreceptors in the eyes. All the cones have a maximum sensitivity, or a different wavelength of light that they are maximally responsive to. For blue cones that is a wavelength of 419 nm (violet), for green cones that is 531 nm (green) and for the red cones they are maximally sensitive to 558 nm (yellow-green). All cones respond to a broad spectrum of light that overlaps the other cones. The final colour experienced is built up by comparing the response from the different types of cones.

1.1.1 CVD Classifications

Acording to Simunovic, ”Congenital colour vision deficiency results from genetic mutations that affect the expres- sion of the full complement of normal cone photoreceptors”

[23]. The mutations are classified in three classes according to their severity: anomalous trichromacy, dichromacy, and monochromacy. Anomalous trichromacy is the least severe version and like normal colour vision it requires three primary colours (all types of cones) to match any colour. The matched colour is what is experienced by the viewer. The colours that are considered a match differ between a person with normal colour vision and a person with anomalous trichromacy CVD. A person with CVD has reduced sensitivity in one or several of their cones that affects how colours are experienced. Within the classification of anomalous trichromacy the impact on the colour vision for different people can vary from just noticeable to dichromacy. The classification is further divided into protanomaly, when it affects the red cones, deuteranomaly when it affects the green cones, and tritanomaly when it affects the blue cones (see Figure 1).

Dichromacy is a more severe form of CVD. Instead of three primary colours to match any colour a person with dichromacy only require two. The classification is subdi- vided into protanopia, deuteranopia and tritanopia [23]. A person with dichromacy has no functional red, green, or blue cones respectively. This has a stronger effect on the colour

(5)

Figure 1: An example of the perceived colours for persons wtth normal colour vision, deuteranopia, deuteranomaly, protanopia, protanomaly, tritanopia, and tritanomaly

vision than anomalous trichromacy. The most severe form of CVD is monochromacy, where the cones are severely dys- functional or non-functional and instead the rods in the eye are responsible for the colour vision. The rods are normally used for night vision and have a very limited capability to match colours. People with monochromacy can often not distinguish between colours and are said to see a monochromatic world, hence the name [23].

The most common form of CVD is red-green, it is a term used to encompass protanomaly, deuteranomaly, protanopia, and deuteranopia. Approximately 75% of those with CVD are green-diminished and most of the remaining are red- diminished [15]. Red-green colour perception impairment is linked to a persons biological sex and caused by a mutation that commonly occurs in the X chromosome. Because men only have one X chromosome they have a higher likelihood of red-green CVD, 8% compared to women who only have 0.5%

likelihood[23]. Red-green CVD results in reds and greens being perceived as shades of yellow (see Figure 1). Therefore, it is hard to distinguish between the two colours.

1.2 Diagnosing CVD

The highest regarded way of detecting if a person has a CVD is to do a diagnosis with anomaloscopes [20]. The Nagel anomaloscope is designed to test for red-green dis- crimination. The test person views a circle where one half is yellow light at 589 nm and the other half mixes red (666nm) and green (549nm) light. The test person then needs to com- bine the red and green light so that it matches the yellow light. With a skilled examiner, this can be used to get a very accurate diagnosis of CVD. Due to that the anomaloscopes present colour as monochromatic light it is very difficult to do a computerized version of the test as a screen uses a com- bination of red, green and blue (RGB) light to display the yellow light on both sides of the circle.

The RGB colour space used in computer monitors is the colour mode that most closely mimics the colour vision in the human eye [15]. Both the eye and the RGB colour space is additive so they will show brighter colours as more light is added. One problem is that computer screens rarely show the same colours as the computer intends to due to the screens calibration [8] and this can be an issue for tests for CVD on screens. With an exact calibratied screen it can

Figure 2: An example of a test plate used in the Ishihara test

Figure 3: Digital implementation of Farnsworth Munsell 100-hue test based on 88 coloured plates

be considered to show absolute colours and be suitable for CVD tests [11].

Two popular tests that have been shown to work well when conducting CVD tests on computer screens are the Ishihara test designed by Dr. Shinobu Ishihara in 1917 [25]

and the Farnsworth Munsell 100-hue test [11]. The Ishi- hara test is based on images built up by circles with varying colours hiding numbers within them designed to be difficult to distinguish with different forms of CVD (see Figure 2).

The Ishihara test is conducted by showing several different images to detect red-green CVD which Tsai et al. [25] show this to be approximately equivalent of an anomaloscope test.

The Farnsworth Munsell 100-hue test consist of 100 coloured plates in a spectrum of colour divided in four rows (see Fig- ure 3). The end plates are fixed on each row and the middle ones should be arranged by the test person as they see fit.

From the amount and position of the deviations between the test person and the gradual transition between hues an accurate diagnosis of CVD can be attained. Ghose et al. [11]

show that this test works even more efficiently on a computer with a calibrated screen than it does in its original physical form. The results are approximately as accurate, the time to conduct the test is shorter, there are no errors when scoring the test, and the results are instant.

(6)

2. THEORY AND RELATED RESEARCH

The problem of not considering colour vision deficiency when developing websites are that there might be information that is not accessible or clear due to the colour that the designer or developer has chosen to use. A common issue that affects people with CVD is caused by not having sufficient contrast ratio between text and its background. This can make the text difficult or impossible to read. This effec- tively excludes a large portion of the people from a website.

Today a common approach to help developers when build- ing websites for people with CVD is to use visualising tools such as Spectrum [17] or Colorblinding [5]. These enable the developer to preview the site as if they themselves were colour vision deficient. These tools are accurate but each page has to be previewed for each of the different types of CVD. A very small website might contain five different pages, and to check them for all eight types of CVD that is 40 pages that has to be evaluated. It is not uncommon that a large website has hundreds or more pages and this makes this evaluation very time consuming, as already mentioned.

Another problem is that the tools require the developer to remember to use them or to have these tests as a routine part of website development. This is not currently standard practice in the software development settings that were examined.

2.1 Adapting colours for people with CVD

There have been several studies conducted on how to adapt colours in images and alter illustrations to better suit people with CVD. These studies show that there are different approaches to the problem. Geissbuehler and Lasser [10] work with creating colour schemes for presenting data and suggest an adaption of what colours to use to display data to red-greed CVD people. As an example, instead of red and green it is suitable to use green and magenta or blue and orange when working with two colour channels.

They also suggest two colour schemes, Isolum and Morgen- stemning, that could replace the rainbow colour scheme in illustrations [10].

2.1.1 Daltonize

Anagnostopoulos et al. [1] used the algorithm developed by the International Telecommunications Union to convert RGB colour space to a so-called LMS system. A LMS system is based on relative excitations of the longwave (L, red), middle wave (M, green), and short wave (S, blue) cones.

This made it practical to calculate the alterations in colour perception for dichromats as they lack one class of cones.

The conversion of the colour space reduces the calculations to a matrix multiplication that can convert colours to the perceived colour for a person with red-green CVD. A similar approach is taken by Huang et al. [12] where they convert RGB to CIEL*a*b (also a LMS colour space) so that they can use the Euclidean distance to calculate the perceptual difference between two colours. Their research reaffirmed previous research by Vi´enot et al. [26] that used a LMS colour space to create replacement colourmaps that allowed designers to visualise how deuteranopes and protanopes experience colours.

The advances made in the mentioned research lead to a popular concept, to convert the normal output colours to a colour scheme that is distinguishable by people with CVD, Daltonizing. One approach to this was researched by Tanaka

et. al [24] who use a lightness modification method to increase the contrast between colours that are hard to distinguish for people with protanopia or deuteranopia with minimal visual effect for people with normal colour vision. The CVD person more clearly experiences the different shades of yellow, and this makes it possible to easier distinguish between different objects that are coloured in difficult to distinguish colours for CVD persons.

Olivera [18] developed another method that changed all colours in an image that was hard to distinguish for people with CVD to another more contrasting colour that made the image easily perceptible. Although this was efficient for visualisations the result was less pleasing as the par- ticipats experienced the images differently. Participants of Oliviers’s test said that ”the colours in these images didn’t match [their] previous experiences and, as such, didn’t look

’natural’” [18].

2.1.2 Colour contrast

The majority of previous research on CVD addresses the issue of colour contrast. The contrasts impact for people can be quantified through using a contrast sensitivity function.

Johnson et al. writes,

The contrast sensitivity function (CSF) measures the sensitivity of the human visual system to linear contrast as a function of spatial frequency.

Sensitivity is defined as the reciprocal of the contrast threshold or as the minimum amount of contrast necessary to elicit a response. [14]

The function can be used to calculate the differences and to decide threshold values for noticeable difference between two colours. Doliotis et al. [9] found out that in their research that the needed colour difference in the RGB colour space was a difference of 10 units in all the respective colour channels.

The calculations in the two previous sections are of great significance, and are the mathematical base that this report builds on.

2.2 Research question

Most of the current tools for developing with CVD in mind are accurate but require that the developer remembers to work actively with them in order for the tools to have any positive effect. This puts all the responsibility on to the developer who is often not an expert in inclusive design nor personally motivated or given proper incentive to take on this responsibility. This research aims to examine the pos- sibility to develop a tool that can move the responsibility of remembering and deciding to design for people with CVD to the tool and have it to inform the developer when ac- tion is needed. To achieve this the tool needs to be able to identify issues related to CVD. In this research project, the focus is on detecting issues for someone with protanopia or deuteranopia as their severity is high enough that it affects their daily life and almost all of CVD people are red-green deficient [15]. Therefore the research question that will be examined is: Can a computer system, such as a web browser extension, reliably detect issues that impact upon internet viewing experience of people with colour vision deficiencies?

(7)

3. METHOD

To be able to test the reliability of a computer system for identifying issues for people with CVD a tool in the form of a web browser extension was developed. The extension was built for the web browser Chrome, made by Google, and was used as a reference and compared against an evaluation carried out with CVD participants.

3.1 Methodology

The report is written from the ethical perspective of inclusive design. The theory of inclusive design is built on a

common shift in thinking was to replace the view that people are disabled by physical and mental impediments with the more radical proposal that people are disabled by designs and environments that do not take account of the full range of human capabilities [6]

This theory is directly applicable for CVD as there is no limitation in the capability level of people with CVD that cannot be compensated for with inclusive design.

3.2 Chrome extension

The Google Chrome extension was built to analyse luminance and colour contrast. It is based on two algorithms, the first one is W3C’s recommended algorithm for relative luminance, see Equation 1 [21] and the second one using the CIEDE2000 [22] standard algorithm implemented by Huber et al. [13] to calculate the colour difference. These are used to calculate if there is enough contrast between colours. The extension has been calibrated and compared to the participants’ test results to get the closest approximation of their colour vision. For later development on a commercial prod- uct a sufficient margin should be selected to ensure that there is enough contrast between colours to not miss any potential issues. W3C[21] has recommended values for this purpose.

L = 0.2126 ∗ R + 0.7152 ∗ G + 0.0722 ∗ B (1)

where L is the relative luminance, R, G, and B is the relative value of the input color’s red, green and blue channels.

The extension parse through the Document Object Model (DOM) and registers all colours. To calculate the values for the colour that simulate the view for protanopes (people with absent red cones) and deuteranopes (people with absent green cones) matrices (2) and (3) are used in linear trans- formations. (2) for protanopes and (3) for deuteranopes.

These were calculated by Vi´enot et al. [26]. After adapting the colours to simulate dichromacy the CIEDE2000 and W3C’s relative luminance algorithms are applied to calculate if there are any issues in contrast between any of the colours. In the case of contrast issues the extension checks if the colours are directly adjacent to each other. If so it is considered as an issue that needs to be addressed, otherwise it is disregarded. All registered issues were presented to the developer using the extension.

Figure 4: A participant doing the F-M 100-hue test







0 2.02344 −2.52581

0 1 0

0 0 1





 (2)







1 0 0

0.494207 0 1.24827

0 0 1





 (3)

3.3 Evaluation

To be able to evaluate the results calculated by the Chrome extension, seven evaluations were conducted. These were used as a reference to see if the issues identified by the Chrome extension during the test cases were identified by actual people with CVD. The participants were recruited through social media and people that were aware of their CVD were targeted. The participants were all male due to the fact that protonomaly and deuteranomaly is genetic and mainly affect males [23]. The evaluations were conducted on a computer screen that was calibrated with a hardware device that measures that the screen’s colour output corre- sponds to the colour intended to be displayed [8]. The evaluation was conducted in a room where the light could be controlled and replicated for each individual evaluation. Made up of two stages, the evaluation started with a Farnsworth Munsell 100-hue test (F-M 100-hue test) [11], to determine the severity and the type of the colour vision deficiency of the participants. This was used to determine if the participant’s result on the evaluation should be compared to the Chrome extension’s calculated values for protanopia, deuteranopia or both. (see Figure 4).

All data was recorded in writing. Originally there was a plan to film the evaluations to be able to examine the results and the process after the evaluation had been conducted. Several participants expressed that they where not comfortable with the filming and that it felt like they where being tested. Because this group can be considered vulnerable [27] and the study focuses on quantitative data and filming mainly focuses on gathering of qualitative data, not being able to film the evaluations had a minimal effect on the data gathered.

The evaluation was conducted by showing ten different test websites with 30 tasks in total to the participants and they would describe what they saw. The test websites were

(8)

Figure 5: a. left: A test site without any contrast difficulties. b. right: A test site with contrast difficulties for both protanope and deuteranope participants

designed by the researcher based on the logic behind the Ishihara plates [25]. The goal was to build several websites that are intentionally difficult for people with red-green CVD to distinguish the font from the background and different coloured boxes. To avoid learning effects and increase the reliability of the results there were also several websites included in the evaluation which did not present any difficulties. A test website without any contrast problems was used to explain how the test would work (see Figure 5a).

A test website that presented contrast difficulties for both protanope and deuteranope participants (see Figure 5b).

As a second reference point a evaluation was conducted with a participant that had close to perfect colour vision (confirmed by a F-M 100-hue test). The aim was to confirm that the results for the Chrome extension for normal colour vision were accurate and to confirm that there was a difference between the vision of a CVD participant and a participant with normal colour vision.

The ten test sites had three parts to them, a large font, a small font (sizes according to W3C guidelines [4]), and a set of boxes (see Figure 5). The participant answered if each font was clearly readable, readable or not readable.

This was translated to 2, 1, and 0 to be able to compare this to the Chrome extension result. These numbers were chosen to indicate if it was a slight difference or a larger difference between the extension and the participants’ result.

The number of boxes that the participant could see was qualified as those which they could see with-out effort, for example if they could see another box when they looked closely and focused hard then it was not counted as a visible box.

3.4 Analysis

When the participants were done, their result was compared to the result of the Chrome extension. If the participant and the chrome extension identified the same issues or both did not identify any issues, this was considered a successful test-case for the extension. If however the participant and the chrome extension identified different issues this was considered a failed test-case. The severity of the failure was determined by the difference between the participant’s and the extension’s result. For example if the par-

ticipant replied 2 (clearly readable) and the extension gave 1 (readable) the difference was 1, if the extension gave 0 (not readable) the difference was 2. The score was calculated as the total difference between the extension’s and the evaluation participant’s result.

The results reliability was analysed by calculating the statistically significant difference using the comparative error formula 4 [2] between each participant and the extension’s calculated result for the participant’s type of CVD. The formula is designed to compare two results and to see if there is a significant difference between them in relation to the amount of data that it is based on. If the comparative error is smaller than the difference between the results in percent then it is considered as statistically significant. If there was no statistically significant difference between the participants’ result and the extension’s calculation for their CVD but there was a statistically significant difference from the normal colour vision, the result was considered to confirm the reliability of the extension. However, if there the results correspond to the normal colour vision and there is a statistically significant difference between the participant’s CVD then the result is considered to contradict the reliability of the extension.

E = 1.96 ∗

rr1(100 − r1)

s1 +r2(100 − r2)

s2 (4)

where E is the comparative error, 1.96 referse to a 95% confidence intervall, s1, s2 and r1, r2 is the first and second sample size and percentage response

4. RESULTS

The results are separated into a two parts: the extension’s results in comparison to the participant’s with almost perfect colour vision (reference value); and the second part which shows the results of the extension compared to the results of one particular participant with CVD. The results of this participant (P1) are used as a representative example because they are statistically significant. For a survey of all the participants’ results see Appendix A.

4.1 Extension vs Reference Value

The extensions performance for normal colour vision was compared to a participant with near perfect colour vision.

The tables below show the result of the extension’s calculation for normal colour vision (see Table 1) compared to the reference value, the evaluation participant with almost perfect colour vision (see Table 2).

Table 1: The test result from the extension’s calculation for a person with normal colour vision. The differences compared to reference participant are marked in orange

Test website 1 2 3 4 5 6 7 8 9 10

Large font 1 2 2 2 2 2 2 2 2 2

Small font 1 1 2 2 1 2 2 0 2 0

Nr of squares 9 7 8 7 5 8 8 6 4 9

(9)

Table 2: The test result from the participant with almost perfect colour vision. The differences compared with the extension is marked in orange

Test website 1 2 3 4 5 6 7 8 9 10

Large font 1 2 2 2 2 2 2 2 2 2

Small font 0 1 2 2 1 2 2 0 2 0

Nr of squares 9 7 8 7 5 8 8 6 5 9

The difference between the calculated result for normal colour vision and the reference participant is one out of 30 tasks. The statistical significance is determined by the comparative error formula 4. Where s1 and s2 is set to 30 (number of tasks in an evaluation) and r1 is set to 100% as the reference participant is considered as the correct answers in this task and r2 is set to 96.7% (29 out of 30 in percent) as there were a difference in one task. The calculations can be seen in 5. As the comparative error is 6.39 and the difference between the results in percent is 3.3 there is not any statistical significant difference at a 95% confidence interval.

Out of the 30 tasks comparing the calculated normal vision to the extension’s CVD calculations, protanopia (red deficiency) differed in seven tasks and deuteranopia (green deficiency) in nine tasks. These have a comparative error at 15.1 with a 23.3 difference in percent for protanopia and a comparative error at 16.4 with a 30 percent difference for deuteranopia. Both these differences are statistically significant compared to the calculated value for the normal colour vision.

6.39 = 1.96 ∗

r100(100 − 100)

30 +96.7(100 − 96.7)

30 (5)

4.2 Extension vs CVD participant

The results from the CVD participants (see Table 3) were also compared to the algorithm’s calculated result for both protanopia see Table 5) and deuteranopia (see Table 6). Ta- ble 3, 5, and 6 are examples of the results from the example participant P1.

Table 3: An example of the test result, showing the result from participant P1

Test website 1 2 3 4 5 6 7 8 9 10

Large font 1 2 2 1 2 2 2 2 2 2

Small font 0 0 2 2 0 2 2 0 2 0

Nr of squares 9 7 8 7 5 7 8 6 4 8

Table 4: The test results from the extension for normal colour vision. The differences between the CVD participant P1 in Table 3 are marked with orange.

Test website 1 2 3 4 5 6 7 8 9 10

Large font 1 2 2 2 2 2 2 2 2 2

Small font 1 1 2 2 1 2 2 0 2 0

Nr of squares 9 7 8 7 5 8 8 6 4 9

Figure 6: Screenshot of the participant’s result on the F-M 100 hue test. It shows the severity and the type of colour vision deficiency

Table 5: The test result from the extension, calculated for protanopia. The differences between CVD participant P1 in Table 3 are marked with orange

Test website 1 2 3 4 5 6 7 8 9 10

Large font 1 2 2 2 2 2 2 2 2 2

Small font 0 0 2 2 1 2 2 0 2 0

Nr of squares 8 7 7 7 5 7 7 6 4 8

Table 6: The test result from the extension, calculated for deuteranopia. The differences between the CVD participant P1 in Table 3 are marked with orange

Test website 1 2 3 4 5 6 7 8 9 10

Large font 1 2 2 2 2 2 2 2 2 2

Small font 0 0 2 2 0 2 2 0 2 0

Nr of squares 9 7 7 7 5 7 7 6 4 8

As can be seen in Tables 4, 5, and 6 comparing the extension’s calculated results to the CVD participant’s results it correlates with the results for deuteranopia (the participant was diagnosed with deuteranopia), as there is no statistical significance between the results at a 95% confidence interval.

The results are not correlating for protanopia, as there are a statistically significant difference. If the result is compared to the reference value in Table 2 or 4, there is a statistically significant difference between the CVD participant and the reference participant (participant with near perfect colour vision) or the calculated value, in colour vision at a 95%

confidence interval. Statistics also illustrate the severity and type of the CVD participant’s condition used in the example results (see Figure 6).

The results and severity for each of the evaluation participants can be seen in Table 7. A severity below 70 is considered mild and above 300 is considered severe according to [7]. The main type of colour vision deficiency is either deuteranopia or protanopia.

As can be seen in Table 7, five out of seven of the participants had a smaller difference between their result and the extension’s calculated result for their type of CVD compared to the difference between the reference participant and the

(10)

Table 7: Summary of the type of CVD, severity, and test results for all participants (P), Normal colour vision (Norm), Protanopia (Protan), and Deutera- nopia (Deutera) is referring to the difference from the extension’s result. The significance (S) indicates if it was a statistically significant result compared to the participants’ type of CVD

P CVD Severity Norm Protan Deutera S

P1 Deutan 212 6 5 3 Yes

P2 Deutan 320 6 5 5 No

P5 Protan 68 4 3 3 Yes

extension’s calculated result for a none CVD person. Three of these had no statistically significant difference from the extension’s calculated result. Two out of seven had results that were closer to the extension’s calculation for the normal colour vision. Six out of seven were classified as deuteranopic (green diminished) CVD and the last one was classified as protanopic (red diminished) CVD.

5. DISCUSSION

The discussion is separated in to three sections: statistical significance, inclusive design, and method criticism. The statistical significance section discusses the results from all participants and the reliability of the extension. In section two the results are evaluated from a inclusive design perspective. The last section discusses the methods potential impact on the results.

5.1 Statistical significance

As can be seen from the calculations based on Equation 4, the extension can be considered accurate at a 95% confidence level for colour contrast for users with normal colour vision. The result for the calculated values for protanopia and deuteranopia gave only three cases (participants P1, P4, and P5) that did not show any statistical significant difference between the results for the calculated value for the participants’ respective CVD and their results’. They did however show a statistically significant difference between their results’ and the reference value indicating that in these cases the calculations from the extension is reliable at a 95%

confidence level.

Participants P2 and P3 did show statistically significant differences when the evaluation results were compared to the calculated values for both normal colour vision and for their CVD vision. Therefore, the calculations cannot be considered reliable at a 95% confidence level. They did show that the calculated values for the participant’s type of CVD was closer to their result than they were to the normal colour vision. This indicates that extension’s calculations are correct but it is not statistically verifiable. That there were no statistically verifiable results was likely caused by a mis- match between the CVD participant’s severity and the extension’s calculated severity (or sensitivity). Vi´enot et al.

[26] note their transformation matrices are for transforming

the colours to mimic the way in which they would be experienced by dichromat, protanopes and deuteranopes. The Farmsworth-Munsell 100-hue test used in the study gives an indication of the severity of the CVD but not the type (anomalous trichromacy, dichromacy, or monochromacy).

This made it difficult to quantify if the evaluation participants have been diagnosed with the same level of severity as the extension is built to mimic. This could also have affected the results of participant P7.

According to the few evaluations carried out the extension seemed most accurate with participants that had a CVD severity in the range between 68 to 212. The three participants (P1, P4, and P5) with statistically significant results can be found in this range. This range also included test person P7, one of the participants with a result that was closer to normal vision rather than the calculated result for their CVD. This presents a slight anomaly and to explain this result more extensive tests would have to be performed to confirm the severity of the participant’s CVD and to make sure the test websites in the evaluation where suitable for that severity and type of CVD.

The most surprising results were from participant P6. The participant outperformed the reference user with close to perfect color vision and identified differences that where designed to be hard to see for a person with normal colour vision. There are several possible explanations for the result, one could be that the evaluation participant’s specific type of CVD helped to enhance the colour contrast to make the different objects more visible. The participant expressed that he was very good at identifying differences in colours and often out performed his friends in games related to colour circulating on social media. This and the researcher’s im- pression from the test hints that a more probable explana- tion to the result is that the participant saw the test as a competition in identifying all boxes and to say that he clearly saw all the fonts rather that saying what was clearly readable, readable or not readable. Therefore the result from participant P6 was considered as less reliable than the other participants.

5.2 Inclusive design

The results indicate that the Chrome extension and its algorithms are able to potentially identify issues related to colour vision deficiency and website accessibility requirements. This could have a significant impact on the web development regarding inclusive design. Compared to the many solutions on the market (such as Spectrum [17] and Colorblinding [5]) that aim to raise the quality of products developed for CVD internet users by allowing the developer to preview what they build, the Chrome extension developed in this study is not as dependent on the developer remembering to use the different tools. If the website development is carried out according to the extension’s recommendations it will be unnoticed in the background, but if the design being implemented is using a colour that make viewing for someone with CVD difficult, the developer will be informed that there is a problem. In this regard it acts as a reminder to conduct inclusive design.

The issue with the calibration of the extension’s sensitivity may not be that important because of the goal with the extension, to enhance the products that are developed to include people with CVD and to increase the awareness of the potential issues. To do so it is important to choose a

(11)

high enough contrast value for what is considered an issue.

If the contrast values are set to the recommended values by the W3C [21] there is a very low risk that any issue will go undetected due to the calibration or the selected threshold values.

If the extension works as a reminder of inclusive design and can help developers to more easily build products for people with CVD, it could have a positive impact on society.

Through including CVD requirements, the extension could lead to more accessible websites. Additionally businesses working with the extension can more easily avoid fines for not developing accessible products for CVD people, and the code developed would be more sustainable as there is less need to rewrite it once the stricter laws are enforced [16].

There is a lot of potential for the extension when focus- ing on accessibility. As the core functionality is to reduce or eliminate issues with low luminance or colour contrast it has the potential to enhance readability and make websites more accessible. This is not only true for people with CVD but for anyone, regardless of their colour vision, as the extension also identifies issues before re-calculating the colours to mimic the vision of deuteranopes and protanopes. The fact that it caters to a larger target group increases the likelihood that the extension will be used by more people. With more users of the extension there will probably be less issues related to low luminance and colour contrast and a higher awareness of issues related to CVD due to the notifications from the extension when unfavourable colour decisions are made. Furthermore, the algorithms implemented in the extension could be applied to several other tools used during the design process of websites and other digital products.

If the extension would for example be added to Sketch [3]

(a popular tool used by designers when designing websites and applications) this would enable them to easier share the responsibility of inclusive design and to try different ideas and to know if they are accessible or not.

5.3 Method criticism

The method was well suited for this kind of research but it had a few uncertainties, the participants, the colour vision deficient test (F-M 100 hue) and the test websites that was used.

5.3.1 Test websites

The test websites were based on the Ishihara plates to only work with well established and tested colours [25]. The selection of the colours from these plates was done by the researcher and depending on which colour was selected, this could give a different focus on the test. For instance if there were a lot of issues related to green deficient rather than red deficient. In this study, colours were selected that would present an equal number of issues between people with red and green colour deficiencies.

5.3.2 F-M 100 hue test

Another thing that potentially affects the analysis of the results is the F-M 100 hue test [7]. As shown by Ghose et al. [11] these types of test work well when carried out on computer screens if they are properly calibrated. In the study the evaluations were conducted on a calibrated screen in a controlled light environment to minimize any uncertainties based on the technology used to conduct the evaluation.

The result of the F-M 100 hue test classified all participants

except one as green deficient, this could be because the tests were biased towards green deficiency or that in fact all, but one of the participants were green deficient. As approximate 75% [15] of people diagnosed with CVD have deuteranopia or deuteranomaly this is likely due to the selection of participants. Regardless of the reason this made it hard to validate if the extension can be considered reliable when considering protanopes.

Another thing that could affect the test results of the F- M 100 hue test is that there is a learning effect and if re- peated the performance of the participant can be increased.

This was likely the case with participant P5. He was actively working to improve his colour vision and was used to do similar exercises. However, this effort to learn occurred outside the bounds of the user test and could not be compensated for with any change in methods or test formation.

This would explain why he seemingly had a quite low severity of CVD but his evaluation results where still closer to the calculations of both protanopia and deuteranopia than they were to the normal colour vision. To avoid learning effects in the evaluation process of the extension, the test websites where designed with several tasks that were either easy or difficult to distinguish between the colours used.

5.3.3 Participants

As the number of participants in the evaluation were few it is hard to draw any statistically reliable conclusions on the result of the evaluation. The study is limited to give indications of a result. These indications are promising regarding calculations for deuteranopia as four out of five participants showed results that better corresponded with the calculated values for deuteranopia than for the normal colour vision.

With a larger number of participants, the study could prove if the extension is reliable or not.

With all male participants, the study is limited to give results relevant for males. As CVD is a physical impairment in the cones, this indicates that the results could be valid for females with the same type of impairment. As the study focus on the participants’ physical abilities this suggests that gender should have minimal effect on the results.

5.4 Future research

One future development in line with this research would be to investigate what impact the severity of the CVD has on the reliability of the extension. As can be seen in the study there is a variation in how accurate the extension is that could be interpreted as connected to the severity of the CVD, but it is not sure that it is the reason.

Another interesting aspect to look at would be to extend the research to include tritanopia and to investigate if it is possible to develop a colour calculation to mimic the type of vision people with tritanopia have, and to check if the extension is reliable or not in these instances.

6. CONCLUSIONS

The results suggests that there are possibilities for reliably identifying issues related to colour vision deficiencies and that a simple web-browser extension can detect and assist in designing web pages that cater for CVD internet users.

The problems related to the calibration of the extension are less important in a real world case as the in goal is not to mimic the exact colour deficiency of the person browsing the internet, but rather allow for enough of a margin so that

(12)

the extension will not miss any CVD-related issues. And thus make sure that all information on the websites is easily accessible. The extension is one critical step towards increasing inclusiveness on the internet through making CVD accessible web pages less time consuming to develop. The extension could also be configured to make sure that websites are displayed with enough contrast for easy readability for all viewers regardless of their colour vision capabilities.

7. REFERENCES

[1] C.-N. Anagnostopoulos, G. Tsekouras,

I. Anagnostopoulos, and C. Kalloniatis. Intelligent modification for the daltonization process of digitized paintings. In International Conference on Computer Vision Systems, 2007.

[2] G. Blom. Sannolikhetsteori och statistikteori med till¨ampningar. Studentlitteratur, 2005. Lund, Sweden.

[3] B. BV. Sketch app. https://www.sketchapp.com, 2017. [Online; accessed 2017-06-19].

[4] B. Caldwell, M. Cooper, L. G. Reid, and

G. Vanderheiden. Web content accessibility guidelines (wcag) 2.0. https:

//www.w3.org/TR/2008/REC-WCAG20-20081211/, 2008. [Online; accessed 2017-02-11].

[5] L. Cardoso. Colorblinding. https://leocardz.com/

become-a-colorblind-in-your-browser-e822677167d5, 2017. [Online; accessed 2017-04-27].

[6] J. Clarkson and R. Coleman. Inclusive design. Journal of Engineering Design, 21(2-3):127–129, 2010.

[7] Color-blindness.com. Farnsworth-munsell 100 huecolor vision test. http://www.color-blindness.com/

farnsworth-munsell-100-hue-color-vision-test/, 2017.

[Online; accessed 2017-04-15].

[8] I. C. Consortium. Display calibration.

http://www.color.org/displaycalibration.xalter, 2017.

[9] P. Doliotis, G. Tsekouras, C.-N. Anagnostopoulos, and V. Athitsos. Intelligent Modification of Colors in Digitized Paintings for Enhancing the Visual Perception of Color-blind Viewers, pages 293–301.

Springer US, Boston, MA, 2009.

[10] M. Geissbuehler and T. Lasser. How to display data by color schemes compatible with red-green color perception deficiencies. Opt. Express, 21(8):9862–9874, Apr 2013.

[11] S. Ghose, T. Parmar, T. Dada, M. Vanathi, and S. Sharma. A new computer-based farnsworth munsell 100-hue test for evaluation of color vision.

International Ophthalmology, 34(4):747–751, 2014.

[12] J. B. Huang, C. S. Chen, T. C. Jen, and S. J. Wang.

Image recolorization for the colorblind. In 2009 IEEE International Conference on Acoustics, Speech and Signal Processing, pages 1161–1164, April 2009.

[13] S. Huber, A. Jenter, and C. Furman. colorlab is a javascript lib to calculate deltae (ciede2000) values between two colors.

https://github.com/signalwerk/colorlab, 2016.

[14] G. M. Johnson, X. Song, E. D. Montag, and M. D.

Fairchild. Derivation of a color space for image color difference measurement. Color Research &

Application, 35(6):387–400, 2010.

[15] D. R. Keene. A review of color blindness for microscopists: Guidelines and tools for accommodating and coping with color vision deficiency. Microscopy and Microanalysis, 21(2):279–289, Apr 2015.

[16] S. Laurin. Nu testas diskrimineringslagen.

http://www.funka.com/design-for-alla/tillganglighet/

tillganglighet-i-diskrimineringslagen--hander-det-nat/

nu-testas-diskrimineringslagen/, 2016. [Online;

accessed 2017-02-15].

[17] Y. Lvivski. Spectrum.

https://github.com/lvivski/spectrum, 2017. [Online;

accessed 2017-04-27].

[18] M. M. Oliveira. Towards more accessible visualizations for color-vision-deficient individuals. Computing in Science Engineering, 15(5):80–87, Sept 2013.

[19] F. OU. Lag (2016:1145) om offentlig upphandling.

http://riksdagen.se/sv/dokument-lagar/dokument/

svensk-forfattningssamling/

lag-20161145-om-offentlig-upphandling sfs-2016-1145, 2016. [Online; accessed 2017-02-15].

[20] G. V. Paramei and D. L. Bimler. Color Vision Testing, pages 407–413. Springer New York, New York, NY, 2016.

[21] C. Ridpath and W. Chisholm. Techniques for accessibility evaluation and repair tools.

https://www.w3.org/TR/AERT\#color-contrast, 2000. [Online; accessed 2017-02-11].

[22] G. Sharma, W. Wu, and E. N. Dalal. The ciede2000 color-difference formula: Implementation notes, supplementary test data, and mathematical observations. Color Research & Application, 30(1):21–30, 2005.

[23] M. P. Simunovic. Colour vision deficiency. Eye, 24(24):747–755, Nov 2009.

[24] G. Tanaka, N. Suetake, and E. Uchino. Lightness modification of color image for protanopia and deuteranopia. Optical Review, 17(1):14–23, 2010.

[25] Y.-C. Tsai, H.-S. Chen, and R. Luo. Screen-based color vision test for digital quantification of deutan color defects using a color-calibrated monitor. Journal of the Society for Information Display, 24(8):487–500, 2016. SID-04-16-0411.R1.

[26] F. Vi´enot, H. Brettel, and J. D. Mollon. Digital video colourmaps for checking the legibility of displays by dichromats. Color Research & Application,

24(4):243–252, 1999.

[27] J. Vines, R. McNaney, S. Lindsay, J. Wallace, and J. McCarthy. Special topic: Designing for and with vulnerable people. interactions, 21(1):44–46, Jan.

2014.