TIC-STH 2009

Image Processing for Colour Blindness Correction

S Poret, R D Dony, S Gregori

School of Engineering
University of Guelph
Guelph, ON Canada
{spicklep|rdony|sgregori}@uoguelph.ca

Abstract— Colour blindness is a genetic mutation that II. BACKGROUND

alters the colour vision of the subjects by decreasing the
A. Trichromatic Vision
sensitivity to certain colour wavelengths, depending on the
defect. There are many forms of colour blindness ranging The human eye contains two types of image receptors: rods
from monochromacy (black-white) to the most common and cones [2]. The rod receptors are sensitive to low light
form, the “red-green” variation where reds or greens are levels but do not differentiate colours. Cones, on the other
weakened, the vibrant shades are easily seen and the dull hand, are only sensitive to brighter light levels but enable us to
shades are difficult to perceive. A filter was designed see different colours. Colour perception is due to presence of
based on the Ishihara colour tests in order to correct the three types of cones: long, medium and short wavelength
colour blind deficiencies. This was successful for seeing cones. They each correspond to a specific light wavelength
the hidden objects within the test plates but did not that represents the three basic colours red (long wavelength),
translate well for real world images. The filter was green (medium) and blue (short) as shown in figure 1. Vision
modified, removing the dullest/lightest shades and shifting that uses three receptors for colour perception is referred to as
all the shades to the darker vibrant shades. The original trichromatic vision.
image was shown to colour blind and normal vision
subjects with results varying among all the subjects. After
the modified filter was applied to a natural image, the
colour blind and normal vision subjects were all able to
correctly identify the test colours.

Keywords-component; Colour blind, image processing,

Ishihara, Red-green, colour correction, filter design

I. INTRODUCTION
Colour blindness is a colour vision deficiency that naturally
occurs within the population. Images with similar colours and
shades can prove to be difficult to view. Object recognition
within images can also be hindered. Therefore a processing
system to aid in this natural mutation would be beneficial. Figure 1. The wavelength of the cone systems [2]
Colour blind individuals often have difficulties in modern
society with traffic lights, paint samples and digital images. B. Colour Blindness
Our focus is on the later: filtering digital images in order to There are three degrees of colour blindness: monochromacy,
correct the colour vision deficiency. There exists little to no dichromacy and anomalous trichromacy [2].
help for most colour blind individuals. Normally this is not Monochromacy is very rare and vision is limited to the
life threatening and most colour blind people live normally. equivalent of a black-and-white movie. Dichromacy is also
Some people may not even know they are affected without rare and is the absence of one of the three cones which causes
testing. Modern societies have even started to aid the colour the total loss of vision of that wavelength. Anomalous
blind with traffic signals; blue hues are added to the green trichromacy is the most common form of colour blindness and
traffic light, while orange hues are added to the red light to simply is the defect in one of the three cone systems. The
further distinguish the three colours from each other. middle/green and long/red wavelength sensitive cones are
However, with modern image processing technology, it may more likely affected resulting in difficulties discriminating
be possible to design an aid to enhance the colour blind’s reds, yellows and greens. This is commonly called “red-
perception of colour in everyday situations. green” colour blindness. Along with the colour difficulties,
duller shades of colours are also harder to distinguish, whereas
vibrant colours are easier to see. The medical names are

978-1-4244-3878-5/09/$25.00 ©2009 IEEE 539

protanomaly for red defect and deutranomaly for the green invisible to those with the defect. Figure 3 shows six
defect. The less common form of trichromacy, tritanomaly, is examples of Ishihara colour plates.
the blue cone defect and results in trouble discerning blues and
yellows.
There are two causes of colour blindness. It can occur in
an accident after birth causing eye, brain or nerve damage.
Most commonly colour blindness is inherited genetically from
mutations on the X-chromosome. Since men have only a
single X-chromosome, men are much more susceptible to
colour blindness: if the single X-chromosome is affected, the
male will be colour blind. If only one of a woman’s X-
chromosomes is affected, she may not present any colour
blindness as the other chromosome could make up for the
defect. Therefore, both X-chromosomes must be affected in
order for a woman to be colour blind. As a result, less than 1%
of women are colour blind whereas 7-10% of males are, [4].
If the woman has a defective chromosome then she will be a
colorblindness carrier. An affected male’s daughters will
typically be a carrier. A carrier’s male children will most
likely be colour blind, depending on if they receive mother’s
X-chromosome or father’s X-chromosome. Figure 2 illustrates
the inheritance rules for colour blindness.

Figure 3. Ishihara plates. From top left: 7, 10, 4, 2, 8, 45

2) Colour Difference Vectors

A quantitative scoring technique for panel tests of color

vision, [5], presents a complex scoring scheme with Color
Difference Vectors (CDV) to determine the type of colour
blind defect, the degree of the defect and the amount of
Figure 2. Inheritance rules for colour blindness [3] randomness in the sample.

C. Tests for Colour Blindness

Over the years, a number of different tests for colour Figure 4. Circle test for CDV analysis
blindness have been developed.

1) Ishihara colour test The circles must be arranged from left-to-right matching the
colour most like the colour directly to the left. Figure 4 shows
The Ishihara colour test, [6] is a standard way to test for “red- the test at starting. Figure 5(a) shows the sorted circles by a
green” colour blindness. It was named after its creator Dr. normal vision subject compared to 5(b) for a colour blind
Shinobu Ishihara in 1917 at the University of Tokyo. The test subject. The patterns within the arrangement of the coloured
has a variety of coloured plates with a circle of dots of circles reveal the estimation of the colour blind qualities.
randomized size and colour. In this dot pattern is a number or
object that should be visible to those with normal vision but

540
 III. METHODS
A. Test Subjects
(a) (b)
First, the CDV method proposed in [5] was used with eight
Figure 5. (a) Normal subject results, (b) Colour blind subject results
test subjects: four colour blind males and four individuals
The arrangement of the coloured circles generates CDVs that with normal vision (two men & two women). The quantitative
are used to create the moment of inertia. The moment of testing proved to be accurate as the test subjects with normal
inertia is yields three important factors: confusion angle (A), vision (NM1, NM2, NF1 NF2) all scored the same (perfect, see
selectivity index (E) and confusion index (F). The confusion Table 2) where the test subjects that were colour blind (CB1,
angle can be used to identify the type of colour defect. The CB2, CB3, CB4) all scored unique results with varied output
selectivity index (major radius/minor radius) tells the amount (Table 3).
of polarity of the subject: the larger the value the more severe
the degree of the vision deficiency. The confusion index is the TABLE II. NORMAL VISION RESULTS
comparison of the normal (perfect) vision arrangement to the
erroneous arrangement and quantifies the colour difference. Results (NM1) Results (NM2) Results (NF1) Results (NF2)
The research in [5] presented a chart for comparison of these A 61.98 61.98 61.98 61.98
values. A large panel of 53 normal and 66 colour deficient B 8.85 8.85 8.85 8.85
subjects were used to calibrate the data and the scoring results C 7.21 7.21 7.21 7.21
of this testing is shown in Table 1. D 11.42 11.42 11.42 11.42
E 1.23 1.23 1.23 1.23
TABLE I. CALIBRATION OF SCORING RESULTS F 0.96 0.96 0.96 0.96

Major Minor Total S- C-

Type of Colour Angle
Radius Radius Error Index Index TABLE III. COLOUR BLIND VISION RESULTS
Vision (A)
(B) (C) (D) (E) (F)
Results (CB1) Results (CB2) Results (CB3) Results (CB4)
Normal 62.0 9.2 6.7 11.4 1.38 1.00
Minor Error -12.1 9.8 9.2 13.4 1.07 1.06 A 11.41 18.73 17.58 13.32

Protanomaly 28.3 18.0 8.2 20.4 1.97 1.95 B 22.32 14.03 16.38 18.39
C 8.97 8.61 8.58 8.45
Protanope 8.8 38.8 6.6 39.4 6.16 4.20
D 24.06 16.46 18.49 20.24
Deuteranomaly -5.8 25.4 9.6 27.5 2.99 2.75
E 2.49 1.63 1.91 2.18
Deuteranope -7.4 37.9 6.3 38.4 6.19 4.10
F 2.42 1.52 1.77 1.99
Tritanomal -80.8 16.3 6.4 17.5 2.57 1.77
Tritanope -82.8 24.0 6.4 24.9 3.94 2.60 All four test subjects tested within the same range of colour
blindness according the study: protanomalous. Test subject
CB1 is shown to have the most colour defect while CB2 has the
D. Previous Work least defect. The results show that all test subjects were within
Little work has been done investigating the use of image the protanomaly level which is the red weakness colour defect,
processing filters to assist colour perception for the colour long wavelength.
blind. Some work has been done in order to attempt to filter
real-world with physical devices. A colour sensing system to It is therefore hypothesized that the red portion of the image
aid the colour blind was presented in [4]. The author designed should be altered in order to aid the colour blind test subjects.
a simple system with two colour sensors that detect the colours Every colour blind individual’s vision is unique and therefore
and sends an analog voltage to a microcontroller. The signal the level of correction may need to be optimized in order to
is then conditioned and related to a colour. The colour achieve accurate results. The Ishihara colour plates will first
detected is output on a display screen where it can be read by be examined and processed in order to determine if colour
the colour blind. The system was effective in performing this correction can aid in the visibility of the numbers in the plates.
basic operation. The wavelengths of each colour were Once this analysis is complete, real world digital images will
detected and the theory behind the anomalous trichromacy be filtered to determine if colour correction can then be
deficiencies was graphed. It was found that the red/green implemented. The processing will be performed in MATLAB.
(common) defects result in the medium (green) and long (red)
curves being shifted to the left. This results in the lack of
detection of these specific wavelengths. IV. RESULTS AND DISCUSSION
A. Preliminary Ishihara Analysis
The Ishihara plates were shown to both sets of test subjects.
The numbers inside the plates (7, 10, 4, 2, 8 and 45) are not
visible to any of the four red-green colour blind subjects

541
whereas all four normal vision subjects could correctly
identify all six values. The designing and testing of various
filters and image enhancement techniques were used on the
images. As a simple approach, the images were processed to
remove the green and blue components leaving only the red
component. This can be easily implemented in MATLAB
using the imadjust command:

A=imadjust(IM,[0 0 0; 1 1 1],[0 0 0; 1 1 1],[1 0 0])

Figure 8. Hidden values with the colour blind filter

More research was to be done around this idea as perhaps

adding in these values helped the colour blind subjects. The
next stage is to apply this approach to real images.
B. Digital Image Analysis
Now, it is known that focusing on the red component itself is
beneficial in colour blind image processing, whereas overall
Figure 6. Ishihara Plate: Red component only. image attributes like contrast and brightness do not play a
large role. Figure 9 illustrates the difference in sensitivity of
By filtering out these two colour components only the “weak”
the red cone receptor. Figure 9(a) shows the normal vision
red component remained. An example Ishihara image is
subject cone sensitivity charts and Figure 9(b) shows the
shown in Figure 6. The image is very bright and the object
protanomalous subject chart. The sensitivity of the read cone
(“7”) within the plate was visible to the protanomalous test
is clearly weakened. The sensitivity of the red cone is reduced
subjects and normal vision subjects.
for all brightness along the curve.
The next filter is built on the same idea of removing the
unaffected colour wavelengths (green/blue) and focus on the
weakened red. The same function was used with the gamma
weight towards the darker (higher) pixels to decrease the
image brightness. In MATLAB, this was accomplished as:
A=imadjust(IM,[0 0 0; 1 1 1],[0 0 0; 1 1 1],[2 0 0])

(a) (b)
Figure 9. (a) Normal vision wavelengths, (b) Protanomaly colour blind
wavelengths [4]

First, we select a test image for enhancement. An image with

(a) (b) lots of colour variety was used as shown in figure 10. The
Figure 7. (a) Gamma weighted towards dark side, (b) Final filter result same technique that was attempted on the Ishihara plates was
used on the detailed “flower” image with the results displayed
As expected the image is slightly darker as seen in Figure 7(a). in figure 11. The image filtered with the preliminarily
The objects within the Ishihara images again are visible to all designed filter adds far too much red to the detailed image.
subjects. The final improvement on the filter was to negate The concentration of the amount of red had to be reduced.
the image while double the red component and removing the The red component added to the original image is weighted
green and blue components. This is displayed in Figure 7(b). towards the darker red shades. This results in the dark red
The light red/white values are all flipped to the reverse end of shades remaining the same and the brighter values becoming
the spectrum resulting in a clear bright object being displayed. darker. This is does not exactly correspond to the shift in
The function used is: sensitivity according to figure 9 as the red wavelength must be
A=imadjust(IM,[0 0 0; 1 1 1],[1 1 1; 0 0 0],[2 0 0]) completely shifted not just the bright values. We now proceed
with this modified filter.
The final generated filtered image was added to the original
Ishihara test plate image. The hidden number within the plate
became practically invisible to the normal vision subjects
whereas some of the colour blind subjects could in fact see the
hidden value in the new image, see Figure 8 (a, b).

542
 weakness and remove these troublesome shades. Once this
was performed the green and blue components of the original
image were re-introduced back into the image resulting in
figure 13. Labels are placed on six test colours of interest (A-
F). The order of the labels (A-F) in the original image and the
filtered image were varied to reduce the chance the colour
blind subject would remember the order of the colours.

Figure 10. Detailed image processing sample with desired colours labeled

Figure 13. Corrected/Filtered “flower” image

The values of all the red components were increased by

19.5%, (50 of 256 total levels). This value was estimated with
the scoring scheme values obtained in the design results.
From the histograms in figure 12; we see that the
brightest/dullest red pixels are all removed (darkened).
Removing the lighter shades of red, which are harder to
distinguish, helps the vision correction process as the darker
more vibrant shades remain. The original image was shown
to five test subjects yielding the results in Table 4:
Figure 11. Ishihara filter applied to “flower” image
TABLE IV. ORIGINAL “FLOWER” IMAGE RESULTS
The histogram of the red component of the original flower     NM1 NF1 CB1 CB2 CB3
image was graphed as shown in figure 12(a). The values of A Purple Pink-Purple Blue Pink-Purple Blue-Purple
the red component were increased by a tested variable amount. B Red Red Red Orange-Red Red
This shifted the values of the histogram to the left, as shown in C Blue Blue-Purple Blue Blue Blue
figure 12(b), darkening the red pixels and effectively D Pink Pink Pink Pink Pink
E Green Green Green Green Green
removing the lightest 50 shades.
F Yellow Yellow Yellow Yellow Orange

The results varied quite a bit. The average colour responses

were incorrect 1/6 or 16.67%. The colour blind subjects had
trouble with the A & B flowers. The filtered image was also
shown to the same four individuals at a different time as
shown in table 5:

TABLE V. FILTERED “FLOWER” IMAGE RESULTS

(a) (b)      NM1 NF1 CB1 CB2 CB3
Figure 12. (a) Original “flower” image histogram, (b) Modified red A Purple Purple Purple Purple Purple
component of the “flower” image B Yellow Yellow Yellow Yellow Yellow
C Blue Blue Blue Blue Blue
The colour blind subjects had difficulties with the weak or D Green Green Green Green Green
E Pink Pink Pink Pink Pink
lighter red values. For this reason increasing the darkness of
F Red Red Red Red Red
all the red values will theoretically compensate for the genetic

543
All five subjects identified the same colours of all six flowers. dullest/lightest shades and shifting all the red shades to the
The colour blind subjects were able to correctly identify the darker vibrant shades. This resulted in an image where colour
previously incorrect colours because of the correction of the blind subjects were able to identify the colours in the image
red component. Referring to Table IV for flower labels; the correctly, while still having the image appear naturally to
correction has removed the blue/pink distortions that are seen those with normal vision. Future research is required to
by the colour blind subjects in flower A, the orange distortion generalize the results in a larger study and to investigate the
seen in flowers B/F and the purple distortion seen in flower C. implementation of such processing in a portable aid for the
The image now appears more vibrant, by removing the weaker colour blind.
shades, and it is much easiest now for the colour blind subjects
to distinguish the different colours.
REFERENCES
V. CONCLUSIONS
Many people are affected by colour blindness, yet little has [1] Blanchet Gerald. Charbit, Maurice. Digital Signal and Image Processing
using MATLAB, Hermes Science Europe. 2001.
been done to investigate image processing methods to aid the
[2] Livingstone, M. Vision and art: the biology of seeing. Harry N. Abrams,
colour blind. This paper has proposed a method to aid the May 2002
colour blind using digital image processing. In the most [3] U.S. National Library of Medicine “X-linked recessive (carrier mother)”
common type of colour blindness, a genetic mutation alters the http://www.nlm.nih.gov/. Retrieved, March 2009
colour vision of the subjects by decreasing the sensitivity to [4] McDowell, Jason. “Design of a color sensing system to aid the color
certain colour wavelengths, depending on the defect. Most blind.” IEEE Potentials, pp 34-39, v 27, no 4, July-Aug 2008.
commonly the “red-green” variation is seen where reds or [5] Vingrys, A.J. and King-Smith, P.E. “A quantitative scoring technique
greens are weakened resulting in vibrant shades being easily for panel tests of a color vision” Investigative Ophthalmology and Visual
Science, v 29, pp 50-63, 1988.
seen and the dull shades not. A filter was designed based on
[6] Ishihara, S. “Tests for colour-blindness” Handaya, Tokyo, Hongo
the Ishihara colour tests in order to correct the colour blind Harukicho, 1917.
deficiencies. This was successful for the test plates but did not [7] Gonzalez, Rafael. Woods, Richard. Eddins, Steven. Digital Image
translate well for real image since there was too much colour Processing Using Matlab. Prentice Hall, 2003.
compensation. The filter was modified, removing the

544