Printable Version of Topic

Posted by: jamescanvin Oct 4 2006, 10:11 PM

A preview of the Duck Bay panorama.

Your all on the right track but not quite there.

To make this I took the thumbnails (64x64 each) from the tracking site and used the tracking data to create a Hugin pan file. (Note the rover quaternian appears a little off in the database at the moment so the horizon isn't flat.)

Then I ran my normal image matching / colouring software on the pan/images to create a full colour mini panorama.



This however has a much lower resolution than the existing navcams, so the next step is to use the pan as a colour layer over the navcam pan (also made using the tracking data so the two match in geometry.)

Then as Fred says, I cut out (poorly, must try harder next time.) just the coloured region from the navcam pan to retain the 'feel' of a pancam pan.



The good news is most of this happens automatically in my software, so you can all expect prompt 'preview' panoramas in the future before the full images are even downlinked from the rover.

James

Posted by: fredk Oct 5 2006, 12:34 AM

Very nice, James. So do I understand correctly that you take the luminance values from the navcam pan, and the hue and saturation values from the scaled up thumbnail pan?

Do you think it would be any improvement to instead do as I originally wrote, namely taking navcam as the red channel, L5 thumbs as green and L7 thumbs as blue (or whatever combination you prefer)? I'm just thinking that these pans are always dominated by the red channel, of course, so could this approach improve the "colour resolution"? Or would this be equivalent to what you're doing?

What's interesting is that if you just tried to superimpose a set of pancams onto a navcam, I think you'd have trouble getting the alignment right since the navcams contain considerable geometrical distortion. Is this accounted for in your approach in generating the projected navcam pan? Or is it negligible at the resolution of the thumbnails anyway? I suppose Michael Howard has sorted all this out with his pancam/navcam combos.

Posted by: mhoward Oct 5 2006, 01:12 AM

Not really. Currently I just match up the Pancams to the Navcams visually. But any distortion in the Navcam optics seems to me to be almost a non-issue, although I admit I haven't looked at it really carefully. The bigger problem seem to be that the Pancams are situated differently on the camera mast from the Navcams, and face a slightly different direction (they have a 1 degree 'toe-in' which the Navcams do not).

By the way, on the a,b,c parameters for the Navcam camera model, I don't know what other folks use but I'm using a=b=c=0.

Posted by: jamescanvin Oct 5 2006, 01:23 AM

That's right, similar to what other have done colouring MI's using pancam images.



I don't think that really improves things, your not really gaining any colour resolution as the green and blue channels that make the colour are still at the lower resolution. And you end up effectively loosing spacial resolution as now two channels containing the low resolution data dominate over the higher red channel.

Another issue is simplicity for me. I have some quite sophisicated routines that i've developed over the last year for making nice colour pans that work on individual pancam pointings (plus knowledge of how they fit together) not on the whole pan. So just adding the green and blue to a red navcam results in some serious colour gradients.

Here is a simple attempt at this showing the problems,





The navcams arn't that bad. I didn't take any into account doing this.

I think the biggest problem is that the navcam isn't in the same position as the pancam so at close range you end up with a small stereo effect. It's more evident in the 'navcams in red channel' attempt in this post (another problem with it!), in fact that image may work as an anaglyph! (Probably the wrong way around, left pancam is to the left of the left navcam so the pancams should be red (i think), you'll have to turn the glasses around. I don't have my glasses to check)

EDIT: Cross post with Mike - glad we agree

James

Posted by: tedstryk Oct 5 2006, 01:28 AM

The thing to do would be to use the thumbs to creat a color image, and then overlay the color image over the Navcam image, which could serve as the greyscale data. I have don e this many times with other data. It can be manipulated to avoid slight parallax. Edit: Looking up the thread, I can see that this is what you originally did. One suggestion...play with saturation levels...because the NAVCAM images the RGB data from pancam don't match entirely, the saturation levels require tweaking to look like Pancam color panoramas.

Posted by: jamescanvin Oct 5 2006, 01:29 AM

I generally use a=b=c=0 but I often optimise for d & e (x & y shift) when doing manual stitching. I couldn't tell you if these come out similar each time or what values thay take however.

James

Posted by: jamescanvin Oct 5 2006, 01:31 AM

That's what I'm doing Ted, see the first post.

Posted by: tedstryk Oct 5 2006, 01:33 AM

Looks like my reply and edit crossed with your reply. Sorry.

Posted by: jamescanvin Oct 5 2006, 01:39 AM

No worries Ted.

Good point about the saturation - I had played with it, but will do some more. I have a few tricks to try out.

James

Posted by: edstrick Oct 5 2006, 08:35 AM

If you composite RGB images at different resolutions together, you not only "muddy" the final resolution, but will get color fringing on sharp, high contrast edges. Ideally, you might have low rez long and short wave frames, combined with a high rez middle-wavelength frame degraded to the same pixel count for color information, and then combine that with the full resolution middle-wavelength data as brightness info to get a full color image.

In early mission images when the links wern't all up and calibrated and the daily data rates were low, They'd transmit (at least from Opportunity) two bands at low resolution and high compression and one band at high resolution and not as severe (though still not low) compression, and you could play smart tricks with that data.

Posted by: tedstryk Oct 5 2006, 02:22 PM

That depends on how much color variation there is in the image. Also, if you are simply trying to create a nice picture, you can manually remove fringing, although it takes forever. The biggest problems are caused by illumination angle changes, as well as the fact that if the grayscale data is from either R, G, or B, or in this case is panchromatic while the RGB data is narrowband, you can get some awkward looking areas.

With these thumbnails, the fringing would be a big issue, but with, for example, Galileo data, when the smaller color frames were 2x2 bins, it wasn't much of a problem except with Europa and sometimes Io.

Posted by: Nix Oct 9 2006, 07:51 PM

question by 'micvoo' on colorizing moved to new thread to keep this one on topic;

http://www.unmannedspaceflight.com/index.php?showtopic=3328