Red sky at night AND morning? Must be Mars!

May 28, 2008 5:47 PM

I want to reconstruct Phoenix Mars lander pics in color. There are five B&W images of the same scene with the wavelength in the jpg headers (445, 485, 533, 604, 753nm). How do I combine these?

There's an explanation and combined color wheel here. Funny thing is, I don't get what they get with GIMP and RGB (604, 533, 485). There's too much red and it looks darker. What am I doing wrong?

posted by jwells to Science & Nature (8 answers total)

Theoretically, all you should need to do is multiply each image by an appropriate RGB color, and add them all together. The problem is that any RGB value will necessarily be an approximation to light of that specific wavelength; your monitor likely uses the sRGB color space, which can represent only a comparatively narrow range of colors. So don't expect to come up with the exact same thing as NASA's "approximate true color" photos.

Also, it's pretty hard to say what you're doing wrong without seeing an example of the output and a description of what you're doing to get it.
posted by teraflop at 6:29 PM on May 28, 2008

This may be what you are doing already, but these b&w images should be loaded as channels into a blank image in GIMP/Photoshop. The "blue" that we know and see with our own eyes encompasses a broader range across the 400s than just 485, so you could try to blend the 445 and 485 to see if the results are a little closer to your liking (which should bring down the red)... Though it's best to prepare the images in an RGB colorspace, the discrepancy with darker image is more likely an artifact of the raw image channels--you can adjust the levels (brightness) across each channel or the image as a whole... I expect that NASA did the same with this image, which falls under the guise of "radiometric correction"
posted by zachxman at 7:08 PM on May 28, 2008

I got something very close to the composite in your link by creating 5 new color channels in Photoshop, and assigning each image to one of the channels. The channel colors were taken using a JPEG of the visible light spectrum, and were pretty crude. One of the images seems offset, but other than that it looks basically like the one in the link.
posted by devilsbrigade at 7:11 PM on May 28, 2008

A funny thing, colour. Just because it looks like there's too much red, doesn't actually mean there is too much red. Try playing with the saturation (decrease it) and lightness (increase it) after the RGB merge (Image - Adjustments - Hue/Saturation in Photoshop)

I got a similar results (flickr image) after an RGB merge (using Photoshop) with 604, 533 and 445, followed by a reduction of the saturation by 26 points and an increase in the lightness by about 11 points.
posted by kisch mokusch at 6:23 AM on May 29, 2008

I like kisch mokusch's image although I feel it is a tad saturated. I'd go that route then mess with the saturation.
posted by Guy_Inamonkeysuit at 7:32 AM on May 29, 2008

That said, goddam, isn't this a sweet mission? GO PHOENIX!!
posted by Guy_Inamonkeysuit at 7:32 AM on May 29, 2008

Response by poster: I've played around a bit with combining (addition, subtraction, division) 485 & 445 and then bringing that in as blue. There's definitely something to the saturation/brightness issue, at least with these four. As I was playing with that I got a chuckle because the top three circles, white, gray, and grayer, is a lot like calibrating a TV using one of the color bars, waiting for the gray lines in the bottom right to be separate colors.

Since the middle one isn't a uniform color like the other ones though, I'm thinking there might be something to devilsbrigade's strategy of bringing all 5 in (juggeling one due to the offset). Even the NASA link I pointed to above isn't unifrorm for that circle, which makes me think RGB is just a quick and dirty look at it without combining all 5. I can't find a way to do 5 in GIMP and don't have Photoshop. Is there another alternative, or technique in GIMP? I'm on a PC, BTW.

Either way, thanks everyone for your help. This is great. I hate waiting for the press conferences where they reveal these things. :-)
posted by jwells at 8:15 AM on May 29, 2008

Best answer: Is there another alternative, or technique in GIMP? I'm on a PC, BTW.

Well, there's imagej (more details here). It's freeware, and a pretty powerful imaging program because of all the plugins that various individuals write for it. The colour merge plugin allows you to merge colour images (one at a time, so it becomes a three step process for 4 channels), which means you can turn each of the b/w images to colour first, and then merge them later. Of course, assigning the RGB values for each wavelength is the tricky bit. There's a (PC) program here that purports to do just such a thing (I can't test it, I'm on a mac), although they warn that "There is no unique one-to-one mapping between wavelength and RGB values", so YMMV.

Good luck, if you get it working it would cool to see a follow-up post of the results!
posted by kisch mokusch at 4:10 PM on May 29, 2008