Quantitation using the Color Mixing Card, revisited

Started 4 months ago | Discussions thread
Flat view
DavidWright2010 Senior Member • Posts: 1,631
Quantitation using the Color Mixing Card, revisited

I am starting a new thread for discussions arising from using a color guide as the test data.

I had previous shown how images of a Color Mixing card ($8 at your local arts shop) might be useful in putting numbers on the perceived differences between Foveon and Bayer rendering of colors, specifically subtle shades of green. I liked this approach since, in theory, one knows the correct RGB values for each grid position. The source colors (left column and bottom row) are known, and presumably RGB values are available. (Ted found the value for patch 6, the magenta color.)

Now, it may be that I need to spend more than $8, and get a better card, but now that I am a little more familiar with ImageJ, there are more ways to analyze this data. As before, I take images of the color card with Bayer and Foveon-sensor cameras:

I applied a 4-pixel Gaussian blur (to reduce noise, and cover for the fact the the Bayer image was not quite in focus) and color-corrected each in PSE (not really necessary since we will be looking at differences, patch to patch.

Then in ImageJ, I split the view into the primary colors, and analyze a line that I drew through any patches I was interested in. Below, I am analyzing row 1 of the Foveon image:

the yellow line shows the data to be analyzed with a profile plot

I did that for both sensors, and see:

The absolute values don't really matter because I will be examining the differences between patches, but it makes me feel better to see similar intensities, so I went back to the color images and adjusted the RGB levels so that they were as close as possible to the full available range of 0-255.

But just looking at the plots above show several interesting points. (I should also mention that ImageJ can write out the list of csv numbers from which it draws these plots, so further analysis of these plots can be done with other programs, or just with Excel programming.) The spikes going upward are from the white spaces between the colors. The flat valleys between spikes are the desired RGB values (in this case the R value) - one can do statistics on the valley values to get a mean and standard deviation. If the valleys aren't flat, there is a problem, usually some glare (this card is glossy finish).

It's very easy to do this. Here are the results for row 1, row 11, row 12, row 13, and column 12. Each grid of graphs below go Blue, Green, Red on the x axis, and Bayer above and Foveon below for the y axis:

row 1

row 11

row 12

row 13

column 12

While there are differences, the two sensors show remarkably similar responses. The most notable difference is the green response for row 1. The differences for the last 6 patches show a much greater change with the Bayer sensor. (Kind of the opposite of what I would have expected...)

OTOH, the first 7 patches in column 12 both less blue intensity and greater blue variation, in the Foveon sensor response. (Actually, as Ted pointed out the sensor+firmware response)

I will repeat these graphs using the method Ted showed for plotting Hue instead of RGB values.

And look into getting a higher quality color card.


 DavidWright2010's gear list:DavidWright2010's gear list
Sigma DP1 Merrill Sigma DP2 Merrill Sigma DP3 Merrill Sigma SD1 Merrill Pentax K-1 +1 more
Flat view
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow