[X] My Assistant

[elakdawalla]

I'm wondering if anybody here has spent any time working on the Tempel 1 images from Deep Impact. The data are available at the Small Bodies Node (see e.g. http://pdssbn.astro.umd.edu/holdings/dif-c...0/dataset.html), and Version 2 of the PDS archive (which appears to have been published 3 years ago) includes versions of the data that have been calibrated, cleaned of instrument artifacts, and converted to units of radiance and also I/F. The HRI data have not been deconvolved to remove the blur: a bit of playing shows that without deconvolution, HRI images look pretty similar to MRI images. Attached is a color composite I made of the highest-resolution full-globe shots of Tempel 1 captured by the MRI before the impact. The image isn't big but there's more variety in color than I expected. This is made from the I/F images and is an RGB composite of red, green, and violet continuum filter images, overlaid over a clear image that I sharpened slightly. It's very red, but I think that's real.

Has anybody else worked with these?

Attached thumbnail(s)

 

[ugordan]

Haven't worked with Tempel 1 data yet because I'm not really inclined to download gigabytes not knowing what I'll find inside, I prefer the PDS Atlas search approach. As for the redness, part of it could be due to the "red" and "violet" filters (the other part would be due to no gamma correction). I'd suggest "orange" and "blue" filters instead and if you really want to split hairs, you could do some channel mixing to bring the wavelengths to sRGB colorspace spec.

The central wavelengths originally chosen for the filters allow for a fairly simple channel mix. Here's what I used on some EPOXI Earth data (OGB filters):

R = 60% red + 40% green
G = 100 % green
B = 80% blue + 20% green

I suspect the very act of replacing the "red" (actually bordering on near IR) with "orange" would give you a more brownish result. Assuming there is such filter data, ofc.

EDIT: Ahh, only now I see the MRI instrument lacks those other filters and only the RED is common.

[ugordan]

Right, here's what playing with RED, GREEN_CONT and VIOLET_CONT channel mixes instead produces:



Your image with the following mixes and brightened 2x:

R = 0.36R + 0.64G
G = 0.11R + 0.89G
B = 0.68G + 0.32B

It would be nice to compare this with HRI ORANGE/GREEN/BLUE filter mixes as in my previous post, if available, especially since these continuum filters are very narrowband. And of course, this mixing business assumes the source spectra is fairly linear, otherwise it starts to fall apart, especially when your source filters are so off-base (like these). The HRI filters are much closer to standard RGB wavelengths and are wideband so aren't likely to miss any spectral features.

[elakdawalla]

Yeah, with the Tempel 1 data especially I wish they had provided tar-balls of just the July 4 data, because that's all that resolves the comet. I used the browse pages to pick out a sequence of interest then just downloaded the few images from that sequence. Here's the relevant browse pages:

MRI: http://pdssbn.astro.umd.edu/holdings/dif-c...5_185/index.htm
HRI: http://pdssbn.astro.umd.edu/holdings/dif-c...5_185/index.htm
ITS: http://pdssbn.astro.umd.edu/holdings/dii-c...5_185/index.htm

Looks like the sharpest full-globe view is from the ITS.

The MRI has very different filters to the HRI, and there's really nothing corresponding exactly to RGB. Attached is a description.

Attached thumbnail(s)

 

[elakdawalla]

Cross-posting Thanks for your version. Guess I'll go dig up those HRI images and see what they produce!

[machi]

Just old images, but nothing perfect.
First is mosaic from ITC, second and third are from HRI.
This treasure still wait for digging.

Attached thumbnail(s)

 

[elakdawalla]

Hooray -- just discovered that IMG2PNG handles these images just fine (Bjorn added FITS conversion when he wanted to use IMG2PNG to open New Horizons files). However the resultant images are mirror-flipped top-for-bottom. To convert these, you don't need to download the LBL files. If you're going to use the I/F files, you need to instruct IMG2PNG to stretch them all the same way. To do that, use this command:

img2png *.fit -ctstretch.txt

And have a file named stretch.txt that contains the following text:

-0.00001 0.1
- -
- -

[tedstryk]

I have worked with a few images.

[elakdawalla]

Nice, is that one of the lookback frames?

Funny story...I was at Von Karman for the DI impact event. During the Q and A session following the briefing where they showed the lookback images for the first time, some random TV reporter chick asked A'Hearn whether he could "explain what was going on in that tunnel image." Everyone on the panel, and everyone in the room, looked confused for a few seconds as they tried to figure out what the hell she was talking about. Then you could feel a mix of contempt and horror wash across the room as everyone realized she was talking about the lookback image, and had had absolutely no idea what she was looking at; trying to make sense of it, her brain had matched it to the view through a dark tunnel. Contempt -- what was she doing in this room, how could she have no idea what she was seeing? And horror, especially on the part of the panel -- does anybody out there in the public have any understanding of what we just accomplished? A'Hearn stammered some sort of basic answer describing that photo. Afterward, I think all the other reporters asking questions worked harder to sound like they were smart and knew what was going on.

--Emily

[tedstryk]

It is a large stack of deconvolved lookback frames.

[elakdawalla]

OK, here's the HRI O-G-B combo. Pretty blurry, and the blur seems to be wavelength-dependent. Doesn't look too different from your modified version of the MRI image, Gordan.

Reason for edit: Corrected channel identification

Attached thumbnail(s)

 

[elakdawalla]

OK, here's a pretty fictional image -- it's the OGB color I posted above, over an ITS frame captured shortly before. There is significant rotation between the HRI and ITS images and I don't know how to deal with that. I'm also attaching the ITS image without adjustment. Anybody got any advice on anything I can do within Photoshop or GIMP to match the two points of view a bit better?

Reason for edit: Corrected channel identification

Attached thumbnail(s)

 

[machi]

Not best, but quickest is probably "per partes" method.
Here is colorised ITC image (color from your O-G-V image).
It's made in Paintshop and slightly improved in GIMP.

Attached thumbnail(s)

 

[ugordan]

OK, here's the HRI O-G-B combo. Pretty blurry, and the blur seems to be wavelength-dependent. Doesn't look too different from your modified version of the MRI image, Gordan.

To be compared in an apples-to-apples comparison to that mixed image, this one should be channel mixed as well (I assume you didn't mix the channels?).
R = 0.6R + 0.4G
B = 0.2G + 0.8B


EDIT: Fixed blue channel mix as it turns out it was in fact blue, not violet channel.

[elakdawalla]

Sorry, Gordan, I meant OGB! I fixed that in my posts above.

Machi, your version looks nice. What do you mean by "per partes"?

[ugordan]

Sorry, Gordan, I meant OGB! I fixed that in my posts above.

I fixed the derived image as well, based on the fact it's a blue channel. The MRI and HRI images now do look pretty darn similar.

[machi]

"per partes". First rough match of the color and BW image is done (color image is resampled and rotated), than misfitting parts are removed and replaced by more suitable resampled version of the color image (mostly same image).
This is simplest and quickest way, when images are similar. More sophisticated is warping, but I don't know, if this feature is in the Photoshop. But It is in ImageJ.

[ugordan]

This is simplest and quickest way, when images are similar.

I do that sometimes as well. Warping the images in Photoshop (or any geometric operation for that matter) lowers image sharpness so I tend to avoid it where I can.

More sophisticated is warping, but I don't know, if this feature is in the Photoshop.

It's available in Photoshop, but (at least in CS2) you don't get many control points so the feature is of limited use when you want tight control like in the example above.

[tedstryk]

I will often work with images at 5x...that significantly reduces the impact of warping.

[ElkGroveDan]

Here is colorised ITC image (color from your O-G-V image).

I could stare at that image all day. Does anyone have any idea of what quality of images we will be seeing from NExT/Stardust? There is a mockup here but my recollection of Wild 2 images was that they were problematic from the volatiles contamination issue which was never entirely resolved.

[elakdawalla]

For context, here's what the raw images from Stardust at Wild 2 looked like.

http://www.planetary.org/explore/topics/as...2_stardust.html

Stardust will get slightly closer to Tempel 1, which is about double the diameter of Wild 2. So in theory we should get better pictures of Tempel 1 than we got from the MRI, but not as sharp as the ITS, and there's no color capability.

[tedstryk]

Stardust's camera is a spare Cassini WAC, so it is quite good.

I don't understand what all this attention to color is about unless we are talking about bringing out extreme minutia to study composition. Based on everything I can tell via HRI, we are dealing with something that is essentially colorless. Other than tweaking out the overall hue of the scene, I don't see much to do.

[ugordan]

I don't understand what all this attention to color is about unless we are talking about bringing out extreme minutia to study composition.

To each his own, I guess.

Based on everything I can tell via HRI, we are dealing with something that is essentially colorless. Other than tweaking out the overall hue of the scene, I don't see much to do.

It's not completely colorless uniform in color - those white patches for example stand out on this hemisphere. For me personally, simple colorizations often don't pass the sniff test in that they look more fake to me than original monochromatic data. In such cases I'd rather have the original grayscale representation.

[tedstryk]

For me personally, simple colorizations often don't pass the sniff test in that they look more fake to me than original monochromatic data. In such cases I'd rather have the original grayscale representation.

That often has a lot to do with poor saturation modeling.

[ElkGroveDan]

That often has a lot to do with poor saturation modeling.

That is definitely true with colorings of images back on the home planet, contrast too. People have a tendency to jack up color contrasts way beyond what is reasonable. With this or any art whether it's writing, or carpentry or anything you spend a long time working on, I find it's best to put the job away for a while then come back a day later and ask yourself what jumped out at you when the project was reopened.

I think Gordan's issue is probably is with "blanket" color efforts where a single color is draped over a large section of an image. A really good colorist (did I just make that up?) will apply subtle hue and saturation variations based on variations in image density that are woven or spotted through a section that might otherwise be treated as one color. There's one person here at UMSF who is a master at it.

[DFinfrock]

A really good colorist (did I just make that up?)

No, you didn't make it up. Colorists have a rich history, going back to those who colored the black and white wood and copper engravings in early printed maps. Antique maps with "original color" are much sought after. Antique black and white maps that have had color added in recent decades are frowned upon.

I'm glad that we can have both grayscale and color versions here on UMSF.

[brellis]

Color pics are really just a relatively narrow band of spectrograph, nest-ce pas?

[tedstryk]

Very true.

[ugordan]

Color pics are really just a relatively narrow band of spectrograph, nest-ce pas?

Not color pics per se, but human eye can see only in some 300 nm wavelength span. Space rocks generally have flat spectra and the curves are not very steep (generally pick your favorite shade of red/brown), providing for very soft hues. If your color pics are composed of say UV and IR filters (in addition to one filter covering the visible - typically green), extending beyond the visible, they are called "enhanced color" in that they effectively increase saturation on abovementioned objects with flat spectra.

[ugordan]

Here are a few color pics I've been working on. Chronologically, first an MRI inbound view showing a subtle shadow the nucleus is casting on its coma (looks better without the white webpage background):



Three views of the impact:


Finally, and probably my favorite is the look-back view, both MRI and HRI imagers:

[Stu]

Some of your most beautiful work yet, without a doubt.

[nprev]

Yes...truly dramatic images with genuine impact (unintentional pun!) Wow.

[ugordan]

Departing Tempel 1 - animated GIF (4 MB) of 12 HRI clear frames taken starting 50 min after impact, 22 000 km distance and ending over 6 hours later and 10x the distance. Shows the impact outburst subsiding over time. Sun is I believe to the left.

[machi]

Back to the archives!
Tempel 1 from DI Impactor's ITS camera.

Edit: "New" image with changed text ((NASA/JPL) -> (NASA/JPL/UMD)) and lower compression.

Attached thumbnail(s)

 

[Hungry4info]

Wow that is nice!

[Decepticon]

OMG!!!

Amazing!

[ilbasso]

With the blurry view off of the focal center of each photo, one gets the visual impression of very shallow depth of field...and without the contexting photos, it's impossible to judge the scale of the alien landscape. You'd be hard-pressed to say that you weren't looking at a close-up of something you were holding in your hand!

[machi]

"With the blurry view off of the focal center of each photo"

It's not "off the focal". I really missed out some more detailed description. It's actually combination of ~50 images taken by Impactor camera, when Impactor approach Tempel 1.
So, every original image has different resolution and different coverage area. Orig. images are combined in some kind of manual superresolution, which provides better results in the central parts of the images, for which more original images exists. Also as Impactor approach comet, image size was changing from 1024×1024 -> 512×512 -> 256×256 -> 128×128 -> 64×64.

[ElkGroveDan]

At first I got the same impression that Ilbasso did, but closer inspection reveals that it is a combination of separate images that just happen to be assembled with the higher res shots in the center to give a macro depth of field impression. If you were to take a potato and photograph it at a 45 degree angle with a wide open aperture, focusing on the center of the potato, this composite is very much what it would look like. If you've ever used a lens-reversing ring to do macro work, this is what everything looks like. Alas, all my examples are on silver and celluloid so I have nothing to share, but this is a good example of what Ilbasso was thinking of.

[nprev]

Yeah, I noticed the same thing IB did as well...very interesting little effect. Thanks for the technical demystification, Dan!

Oh, and of course really outstanding work on the images, Machi, as per your usual!

[djellison]

Reminds me of this image
http://cache.boston.com/universal/site_gra...23_16540139.jpg

Awesome model racing car setup, right?

Nope - just tilt-shift DOF http://www.boston.com/bigpicture/2008/09/t...grand_prix.html