MSL "Drive, drive, drive" toward Glenelg, The scientists (mostly) get the keys - sols 38-56 Options

[fredk]

I was about to write a post saying "I should've trusted you, Emily, I also see nothing in the sol 41 navcams."

Then I saw it. This is extremely subtle - way harder to see than the DDs we saw at Gusev. You can't see it at all when you look at individual frames, just when you animate very quickly.

Here's my animation. I started with an average of all 32 frames, then differenced each frame from the average. Then an extreme contrast stretch, followed by a blur:

There's so much noise here that you have to be careful about seeing random blops of noise "moving", like what can happen when you stare at an empty channel on an analog TV. Still, it's clear that there's a linear feature moving steadily from right to left (roughly east to west).

[elakdawalla]

Yeah, I'd call that subtle all right. I do see what you're talking about, but it's on the edge of credibility. Well spotted. I think I probably won't put this in my post. Hopefully it'll look better in the PDS data.

[Deimos]

Maybe there's a reason there was no movie shown, even given the extreme stretches that were used . I wouldn't get your hopes up about the PDS data in this respect. Easier to pull out--probably. Substantially better product--hmm. I do believe that to be a dust devil: I'm convinced it is not one of the artifacts we sometimes see in NCAM. But that's about all I can say for it.

[fredk]

This is interesting. Since the individual difference frames are so noisy, I thought stacking them in adjacent pairs or quadruplets before making an animation would increase the S/N for the DD. When I tried that, it wasn't much easier to see the DD than in the version I posted above (that's all 32 frames, no stacking). What I think is going on is that when you look at the 32 frame movie, your eyes aren't resolving the individual frames, instead you're blending them together, and effectively stacking some number of adjacent frames. That's why the DD is only visible when the animation runs fast enough. So explicitly stacking before animating doesn't make much of a difference.

Still, if we register all 32 difference frames on the alleged DD (the motion is very close to 1 px per frame), and then stack (average) all the frames, followed by a Gaussian blur and a contrast stretch, we can pretty clearly see the thing:

I think that's about as much as I can squeeze out of these frames!

[Syrinx]

I'm not sure if that's a DD or a Yeti. Or nothing at all.

[elakdawalla]

I think that's about as much as I can squeeze out of these frames!

Nice work.

[utah66]

Fred,

Is this called "focus stacking"? See Wikpedia.org/wiki/Main_Page for today. They show a system for increasing the amount of focused information available in a multi image situation. I guess you were going more for noise elimination that for focus enhancement.

Pat

[fredk]

Yeah, this was just to increase signal to noise. Averaging the frames reduces the noise, because the noise fluctuates randomly from frame to frame so it tends to cancel out. But averaging reinforces the signal (DD) relative to the noise, because the DD is at the same place in each frame (once the frames are shifted to allow for the motion of the DD).

Focus stacking is exactly what mahli does in closeup work.

[RoverDriver]

And that's what Oppy does with all those multiple images taken during an MI stack. At the best focus and layers above and below we take 4 images total (typically, sometimes there is no time or volume) which are then processed on teh ground to increase SNR.

Paolo

[Ant103]

I wanted to do something like for weeks, so, took my mouse, Hugin & Gimp, and created this postcard from Sol 50, in order to have a much bigger sky than the original pano .

[Gerald]

Fred, I see a chance to further improve your dd images (first visualization of a dd during the Curiosity mission, great!) by exploiting a-priori knowledge about the optical properties of the dd.
Assessing horizontal structures as more important (high image frequency) as vertical ones (low frequency), a preferred averaging along the dd axis may enhance the horizontal dd structure while assuming vertical homogeneity.
A first try to investigate this may be an elliptical Gauss blurr (significantly higher than wide) rather than a circular one.
A bit more advanced may be an averaging, parallel to the dd axis, as precisely as possible.

Integration (i.e. averaging) along dd axis might return a grey value for each horizontal position and for each frame, resulting in a moving peak.
That peak may be aligned and averaged (over frames) to get an average peak. That average peak (after subtracting background) may be used to define a specialized edge (or feature) enhancement filter (for vertical edges the shape of the dd) to better pin down the precise dd movement.

May be, finally, we come a bit closer to automatic detection of faint dds.

[fredk]

Interesting first post, Gerald.

I actually had tried smoothing with a roughly elliptical kernal aligned along the DD axis. The results weren't dramatically better, and I was a bit uncomfortable with targeting the processing to make the thing look like I expected a DD to. Although, of course, I assumed it moved like I expected a DD to in my stacking. Still, I'm not exactly trying to publish a paper here, so no real need to worry too much about questionable procedures.

Cool idea to integrate along the axis (ie stack horizontal rows after registering along the axis). That would definitely give more S/N.

The DDs we saw at Gusev were pretty varied, so I don't know how well a specialized feature filter would work. Definitely worth thinking more about, though.

[Ant103]

Sorry for the Back-To-The-Future bump, but I just wanted to show my last "Postcard" production. This is a Navcam, colorized with the help of Mastcam pictures. This panorama was taken at Sol 43, when Curiosity had met Jake Matijeivic.

Enjoy

[djellison]

Now THAT is how you colorize a Navcam pan. That's stunning.

[Tesheiner]

Incredible, Ant.
Even after you stating that is a navcam colorization, my brain still can't admit that. It is perfect!