An Examination Of NASA's Color Methods
How to assemble a color image, and how to undo the errors in the NASA data
The method of assembly is pretty simple. Once you have three images, you overlay them and they blend into a single color image. Each color frame is 1/3 of the image data, and using a "transparency" command lets you determine how much of each ends up in the finished product. But there are some rules that we must follow and some details we must know first.
For one thing, each filter subtracts most of the light from the camera when it is used. Only the particular color that the filter will permit can register in the final image. This means that the image will be very dark, not at all like what you would expect. And, since some colors may contribute more, their images will be intrinsically brighter.
The term for this is "relative intensity". The camera has an electronic aperture and it will adjust itself so that whatever the light level is, it will be "stretched" to cover the entire range of available intensities. This has the advantage of saving the maximum amount of light intensity data in the frame, which means that the final image has potentially more information in it.
However, we cannot just paste the images together- without that relative intensity data, the color balance will be all wrong. So some simple rules of thumb have been developed to help replace the unknown relative intensity data. Calibration targets can be useful for this, as the original colors were imaged on Earth before the rovers were launched. Comparing the digital images that the rover returns to the real images here on Earth, we can get a very good idea of the proper intensity values.
Let's try an image first with no relative intensity adjustment and see how it comes out.
| If no corrections for
relative intensity are made, then all three color frames are given equal
intensity and the image comes out like this. Somehow, while there is
too much green and blue, the image actually comes out more red. This
is in part due to the non-linear nature of our eyes.
This frame was assembled by overlaying the previous frames and then raising the brightness and contrast by 70% and dropping the saturation by 30%. Otherwise, the image are far too dark and the colors are far too rich. Click the image for a full-sized version. |
 |
|
| Now we have adjusted the
relative intensity levels to allow 100% of the red, 60% of the green, and
30% of the blue into the image.
The intensity overall goes down somewhat, so we must raise the brightness and contrast by 75% instead of 70%. The saturation was also decreased by 30% in this image, just as above. The result is a sky that is no longer pink. Note that this is the empirically determined relative intensity value, and the same levels that virtually all photo post-processors use on the NASA data to create the truest color image that they can. The problem is that this image is based on using infrared data in place of the red data. For a true color image, NASA should have shot this image using the 673 nm filter, which is in the red band. Click the image for a full-sized color version. |
 |
|
| Now, how could we verify
whether this method is correct or not? We could benefit from having
a color reference of some sort. And we do, in fact.
NASA included a "sundial" and color reference that allows them to get a very good idea of what things should look like. This image is using the same filters as the above images- L2 as red, L5 as green, and L6 as blue. Here is the result. See the hot pink color chip at the far right lower corner? That is the blue color reference chip. How did it get to be pink? The answer is simple- this image uses infrared as the red channel, not red! The blue reference chip happens to radiate strongly in the infrared band, thus throwing the color way off. It also makes some features of the landscape look far redder than they should. |
 |
|
| In this reference image,
also from Sol 001, I have selected filters L3, L5, and L6 as red, green
and blue respectively. This means that the blue color chip will not
have the excess red added to it. And, I have used the rule of thumb-
red at 100%, green at 60%, and blue at 30%.
Now, it becomes clear that there is not enough blue in this image. And, the saturation is also off. Let's experiment with the levels until we get a true color representation, then apply those resulting values to the original image to get a real idea of what it should truly look like. |
 |
|
| If we remove the rule of
thumb and simply add the colors together at the same levels, the results
are like this- too blue or violet looking. While the heat shrink
tubing on the wiring is almost the proper blue, and the color tabs are
almost as they should be, there is still something not quite right with
the overall image.
So we do have to apply some correction factors. Let's take the values to a midpoint of our rule of thumb and work from there. |
 |
|
| The final analysis reveals
that the sundial and reference chips are not in direct sunlight- they are
being indirectly lit by light bouncing off the landing airbags.
The best balance and correction (without washing out the image) ends up being slightly too green. However, we can compensate somewhat and in the end we get this image. The settings are 78% red, 60% green, and 37% blue. The brightness and contrast are set to +80% and the saturation is down by 27%. And while it still is not quite right, on later images we have a clear shot of raw sunlight to work with and the images are more reliable overall. |
 |
|
| Now, let's see what the
landscape would look like if we apply these new values to the image data.
This image uses exactly the same images as the original L2, L5, and L6 landscape frames, but with one difference. I assembled the image from the data derived from the calibration image above. Now we see (given that we are using the L2 filter instead of L3) that the landscape is almost the same, but the sky is faintly blue. The pink cast is located at the corners, low over the ground. And that is precisely where the dust would be settling in the atmosphere. Since the air is very thin, dust particles will settle much more rapidly than they would on Earth. Now, can we see other evidence of a blue sky on Mars? And more importantly, can we trust that these are actually good calibration values? |
 |
Here is a link to a forum that shows exactly what I have been demonstrating here.