Vischeck logo  


spacer
spacer
User quotes:
Just thought I'd say thanks for providing a great resource.
-Andy (Brighton, UK)
 
Web Vischeck
Wikipedia Affiliate Button

Color blind image correction

(Run Daltonize online here)

One of the most common questions that we are asked is:

Is there a way to cure color blindness?

From a medical point of view, the answer is 'not really'. There are some devices (colored contact lenses or filter spectacles) that are claimed to improve the vision of color blind people but in general, people report very mixed results with them. There is currently no effective way to recover full trichromatic vision if you are red/green colorblind.

So in one sense, color blindness is 'incurable'. But a deeper question is "what is color good for?". Apart from its aethetic appeal, seeing many different colors allows us to distinguish things in the world. Take the example below.


How many types of fruit is this man selling?

People with three different cone types ("trichromats") can see 5 distinct piles of fruit on the ground.

From back to front, they look 'red', 'green', 'red', 'yellow' and 'light orange'.

There are also some baskets on the shelf containing three piles of fruit.


Full color fruit stall
But people with red/green color blindness usually have trouble seeing all the different fruits.
They see something that can be simulated like this:


Note that there now seem to be only three or four different types of fruit on the ground.

It is also hard to tell how many distinct piles of fruit there are on the shelves.
Deuteranope simulation of fruit stall


In some pictures, a lot of important information is conveyed by variations in reds and greens. This is a real problem for color blind people who will miss this information. The important point to understand is that the 'true' color of something may be irrelevant but the fact that it is different from its surroundings is very important.

Here is another example:

Fluorescent labelled cells
Deuteranope simulation of fluorescent cells
1) Cells labelled with three types of fluorescent markers to show up three different structures.
2) Red/green color blind simulation of fluorescent cell picture.

Here, a color blind scientist would be unable to tell the red and green labels apart. They would miss important information about the structure of the cells.

How to fix it?


We can use image processing techniques to improve this situation. There are two ways that we can make information in pictures available to color blind people.

1) The simplest way is simply to increase the red/green contrast in the image. Many color blind people have some residual red/green discrimination. Increasing the red/green contrast makes them more likely to see these types of color variations.

2) We can analyze the information conveyed by variations in the red/green direction and convert these into changes in brightness and/or blue/yellow coloration.  This allows us to map information from a color dimension that is invisible to dichromats into those that they can see.

We call the combination of these two processes 'Daltonization' after John Dalton, the British scientist who was one of the first people to investigate color blindness.

The result of applying these algorithms is shown below.:

Daltonized version of fluorescent cells.
Dichromat simulation of daltonized color cells.
3) Daltonized version of three-color-labelled fluorescent cells.
4) Dichromat (red/green color blind) simulation of the  Daltonized image.

Here, the algorithm has mapped changes in the reds and greens into slight changes in brightness and bluishness. The result is that when viewed by someone with color blindness, the different labelled structures in the cell are clearly visible.

Here are some more examples. In each case, the images are numbered 1 to 4 where 1) is the original full-color image, 2) is a simulation of the color-blind view of the full color image, 3) is the Daltonized version of the full color image and 4) is the color blind simulation of the Daltonized image showing improved red/green discrimination. To see how well the Daltonization algorithm works for someone with color blindness, compare images 2) and 4).

Example: Gauguin painting
1) Original painting
2) Dichromat version of painting
3) Daltonized version of painting
4) Dichromat simulation of Daltonized image
 

In this case, the Daltonization transformation must be judged on a more aesthetic level. While there is more 'information' present in image 4) compared to image 2) the disruption in the color balance may be unsettling (or fascinating) to some dichromats.

Here is an Ishihara plate - a common aid for diagnosing color blindness. The number '45' in the center of the disk is invisible to  someone with red/green color blindness.

1) Standard Ishihara plate
2) Dichromat simulation of Ishihara plate
3) Daltonized version of the Ishihara plate
4) Deuteranope version of Daltonized Ishihara plate
 
There are several things to note about this example. First of all, although the enhancement of the red/green information in the Daltonized version is striking, this is not a good general test of an algorithm to modify images for color blind observers. The reason is that the Ishihara plates are (deliberately) artificial images, and do not represent the range of colors that people see in everyday life. It would be rather easy to design an algorithm to 'improve' Ishihara plates (for example one that replaced all reds with blues) that would nevertheless destroy other types of information in the image. One nice feature of the Daltonization algorithm is that it analyses the information content of the image beforehand and attempts to preserve existing color variations.

Note also that the Daltonizing algorithm does not enable people to "pass" the Ishihara test in any way. It is a digital image processing technique that can make print or video display more salient to color blind people without distorting the color balance to an unnacceptable degree.

And finally, here's the fruit market we showed you at the start:
1) Market stall
2) deuternaope image of market stall
3) Daltonized image of market stall
4) Deuteranope image of Daltnized market stall

The different piles of fruit on the floor are now clearly visible in image 4).

What is it good for? We envisage a number of applications for this algorithm. Here are some examples:

As part of a digital microscope (see the 'cells' example above): Color blind users could flick a switch to improve the visibility of red/green fluorescent labels or intrinsic color contrast...

In digital video recorders and display systems: Viewers could turn on the Daltonize algorithm much as they might vary the brightness or volume using a remote control. Operating on a real-time video stream, the algorithm could improve the appearence of TV for color blind viewers (for example, making sports teams playing in red strip on a green field far more visible) without significantly disturbing the color balance for other members of the audience....

Computer display devices: Computer users could turn on the Daltonize algorithm from as part of their computer video card controls. All output appearing on the monitor (including web-pages, videos and still images) would automatically be processed and rendered more salient to dichromats.

Print media: Images can be Daltonized before printing. Print applications where legibility is essential (for example, public safety documents, maps, technical instructions) can all be made more legible for color-blind users....

That's great! I'd like to use this algorithm.


You can play with the algorithm online on this web page. Please be aware that this page is taking a lot of hits at the moment and may seem a little slow to respond. If you have any particularly interesting images that you would like us to process, send them to us by email at info@vischeck.com and we will run them 'by hand'. Please keep them to a reasonable size :)

We would also like to hear your feedback about this algorithm - especially if you are color blind. Let us know what you think!

If you would like to license the Daltonizing algorithm for use in your own applications, please contact us at daltonize_licence@vischeck.com.


Privacy policy. Contact: info@vischeck.com. Last modified 2012-Feb-01 09:42 GMT.