Color Appearance Simulation

Let's analyze this question.  Suppose someone says: "The green leaves of a tree look yellow to a dog."  He or she must know the definition of "yellow" for dogs.  There is no definition of any color for any animal. Colors are not physical properties of objects.  They are senses or perceptions associated with optical properties of objects.  The association varies among animals. The sense of color is unique.  It is different from most other senses such as pain.  On one hand it is like the olfactory sense - what smells foul to us may be fragrant to some animals.  On the other hand it is different because we can have the same definition of foul and fragrant for all animals and human beings because they are associated with certain types of behavior, while colors are not associated with the same behavior for all creatures.  Anyone can tell what stinks to animals if that smell drives the animal away, but no one can tell what looks red, yellow, green and so on by just observing some behaviors of an animal.  If we discuss this further, we will unavoidably run into psychological, semantic, or even philosophic issues.

The ideal method to visualize the color world of an animal is to create the animal's version of a CIE chromaticity diagram.  The CIE chromaticity diagram was created by doing extensive color matching psychophysical experiments.  If we could do the experiments to get the animal's version of CIE chromaticity diagram., then we could compare it with human's CIE chromaticity diagram  to assign colors (actually names) to the animal's CIE color chart.  Since animals can not speak to us, anyone can imagine how difficult the experiments would be. 

There is no published formal method to visualize any animal's color world.  Many assumptions must be made to do this and many of the assumptions are unfortunately debatable.   

In the author's dissertation done five years ago, the following assumption is made:

The colors we would see if we replaced our visual pigments with that of an animal are equivalent to the colors seen by that specific animal.

In other words, the difference in color perception among animals is caused by the difference in their visual pigment composition. Color depends only on the outputs of cones, or put it in another way, the outputs of cones and colors have a one-to-one mapping relationship universal to all animals.  If we figure out this relationship in humans, then we can apply this to other animals.  This deals with only the animals that are also trichromats.  Simulating the color perception of animals with more than three types of cones are probably beyond our capability. Similarly we can assume what dichromatic animals see is equivalent to dichromatic humans would see if their photopigments are replaced with the two of the dichromatic animal.

A simpler question: what do colors look like to human dichromats?

In other words, how can we trichromats see the world like dichromats do?

Many people have investigated this.  The latest work (Brettel et al, 1997) on this has drawn quite some interest.  It presents an algorithm to transform CRT colors for trichromats so that the output colors simulate the appearance to dichromats. Same as other similar work, the algorithm is largely based on the data gathered from unilateral dichromats.  This is the strongest part of the foundation of the algorithm.  Although this is probably the best work in this category, I have many questions regarding the paper and have been seeking answers from different sources with little success:

1. The algorithm depends on the spectral power distributions of the CRT used by the study (Hitachi Model CM2086A3SG). Theoretically, even if all CRT's comply with the international standard (i.e. produces the same RGB->CIEXYZ matrix), their RGB phosphors can have different spectral characteristics (e.g. different shapes of spectral radiance).  This means the algorithm cannot be used on any other models of CRT without checking their RGB spectral power distributions.  However this algorithm has been probably implemented in many places (e.g. www.vischeck.com) without restricting the selection of the computer monitors.  How good is the simulation if this algorithm is applied to popular computer monitors? 

2. According to the algorithm and Fig. 3 in the paper, there are colors that can be produced by the CRT (i.e. the colors in the parallelepiped) that are projected beyond the wings along a LMS axis.  This is briefly addressed in the discussion section of the paper and it is implied that the trichromats can see all the colors that dichromats can see.  Therefore another question follows logically:  How would those colors be mapped or simulated? 

3. The key part of the algorithm is the two wings in the LMS 3-D space for each type of dichromat.  Although the concept of the wings is intuitive to certain degree,  the paper does not give the rationale behind the wings clearly.  Is there a theoretical or experimental foundation for the wings?

4. Two dichromats (one protanope and one deuteranope) were used to verify the simulation.  The description of the procedure and the results is very sketchy. Since dichromats constitute one to two percent of the population, it is logical to ask:  Are there any more data using dichromats verifying the simulation? Obviously, the best validation of the algorithm is to use unilateral dichromats to check the simulation however this kind of subjects are hard to find.

References:

Brettel, H. et al. (1997) Computerized simulation of color appearance for dichromats. J. Opt. Soc. Am. A 14, 2647-2655.

(originally written on 3/2/2002)