Tuesday, August 17, 2010

Color Coding

The journal "Nature Methods" has a 1-pager in the August 2010 issue (see [1] below for access comments) on the use of color to present data and some of the challenges. Most of us are familiar with FEA output that shows stress/strain/temperature or whatever in a 3-D illustration, with various colors equated to various values of the variables. While the author of the article has his own criticism of some practices in using color, I think most of those are weak [2] and he misses the real issues.

To me, the critical issue that needs to be addressed in every case is whether or not the color scale corresponds to the numeric scale of interest. Too many software packages just autoscale the output to the ROYGBIV scale and call it a day. Maybe you are lucky and it works, but in most cases, I've found it necessary to have the color scale adjusted (often multiple times) so as to focus on a subrange of the variables (perhaps to clarify the coloring), or to use a logarithmic scale or some other issue. (I can recall one case where a dry run for an isothermal slab showed a non-isothermal asymmetry. Wethink this occurred because the FEA was started in one corner and was accumulating errors as it moved across the slab, all of which were very small even in aggregate, but when the autoscale coloring was completed, there they were. Upon rescaling, they of course disappeared.)

One valid point made by the author is addressing some of the optical illusions that can occur. We all seen examples were a color can look to be different if the neighboring colors are different. The approach suggested is to not only scale the colors, but also the saturation of the colors - basically use the gray scale as a second dimension. An added advantage is if the output is later printed in B & W instead of color, all the information is not lost. It's a good idea, one that I will see if I can get our FEA guy [3] to buy into.

[1] Access to this article is quite strange. A subscription to the journal is free, but if you want to look at any article online, then there is a pay-per-view charge. So plan ahead?

[2] His issues are 1) is yellow a lower value than blue? 2) using ROYGBIV as an ordering scheme makes it difficult to pick out intermediate values 3) people can have a number of different perception problems with colors such as occur in optical illusions. #1) is easily addressed with a legend, #2) I will indirectly address above, and #3) is a valid point which the author addresses well (and I refer to in the body of the post)

[3] FEA (and all computer modeling for that matter) seem to me to be something that you either do or do not. If you try and do it part time, you will be a miserable failure. For me, I do not.

No comments: