De-parallaxing color space probe images
Feb. 13th, 2005 11:48 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
mmcirvinI just made a better version of my December 14 Dione flyby composite for Wikipedia.
The first version of this was my first attempt at making a Cassini color composite from JPL's raw image archive. Dione was perhaps an unfortunate place to begin because this picture turned out to be one of the hardest. This is because the Dione flyby was quite rapid, and in between all the exposures Cassini took through different colored filters, the spacecraft moved relative to the moon enough to shift the perspective noticeably. So it's impossible to get the frames to match just by superimposing them; if you get rid of the rainbow fringes in one part of the picture, they'll show up somewhere else.
The right way to do it would be to use map projection software to reproject the frames onto a common spherical coordinate system, but even if I had the software, I'd need precise position data for the moon and spacecraft in order to do that right. So the question is how well I can do by just messing around manually in an image editor. Cassini is set to make an extremely rapid and very close pass by Enceladus in a few days, and if it takes any color separations they'll have even worse parallax problems, so I thought I'd practice by revisiting Dione (and help the Wikipedia article in the process).
I'd managed to make an OK version of this picture back in December, mostly by doing a sneaky trick. I put IR, green and UV shots from the JPL archive into three layers in GIMP, colored them red, green and blue and combined them with an addition rule. Then I slid them around until the craters in the center of the image matched up well, leaving bad color fringes around the horizon.
Then the sneaky trick: I superimposed a flattened version of the resulting composite on a single clear-filter black-and-white image, using only the hue/saturation information in the composite layer to colorize the picture. My reasoning was that the human brain mostly gets texture information from brightness, and color perception tends to be processed at lower resolution anyway. That reduced considerably the blur that otherwise would have resulted from the motion parallax. But you can still tell that the craters near the horizon are kind of discolored.
Today I did the Wikipedia version with a refinement of the same technique. Only this time, I saved eight different versions of the IR/green/UV composite with the layers shifted to optimize the registration in different parts of the picture. Then, for each, I cut out a fuzzy selection around the optimization point, and put all of those into the image as colorization layers covering different parts of the moon.
Also, the rightmost part of the image had some data dropouts on every other scan line (I'm not sure about this, but I think these come from Cassini's data compression scheme when the planners underestimate the data allocation needed for the picture). In December I just hacked out that part. This time I got rid of the dropouts with a video deinterlacing filter, then used a piece of that version as a patch to cover over the damaged part of the main image.
The resulting image still has rainbow fringes in it if you look closely, but I'm getting better at this.
The first version of this was my first attempt at making a Cassini color composite from JPL's raw image archive. Dione was perhaps an unfortunate place to begin because this picture turned out to be one of the hardest. This is because the Dione flyby was quite rapid, and in between all the exposures Cassini took through different colored filters, the spacecraft moved relative to the moon enough to shift the perspective noticeably. So it's impossible to get the frames to match just by superimposing them; if you get rid of the rainbow fringes in one part of the picture, they'll show up somewhere else.
The right way to do it would be to use map projection software to reproject the frames onto a common spherical coordinate system, but even if I had the software, I'd need precise position data for the moon and spacecraft in order to do that right. So the question is how well I can do by just messing around manually in an image editor. Cassini is set to make an extremely rapid and very close pass by Enceladus in a few days, and if it takes any color separations they'll have even worse parallax problems, so I thought I'd practice by revisiting Dione (and help the Wikipedia article in the process).
I'd managed to make an OK version of this picture back in December, mostly by doing a sneaky trick. I put IR, green and UV shots from the JPL archive into three layers in GIMP, colored them red, green and blue and combined them with an addition rule. Then I slid them around until the craters in the center of the image matched up well, leaving bad color fringes around the horizon.
Then the sneaky trick: I superimposed a flattened version of the resulting composite on a single clear-filter black-and-white image, using only the hue/saturation information in the composite layer to colorize the picture. My reasoning was that the human brain mostly gets texture information from brightness, and color perception tends to be processed at lower resolution anyway. That reduced considerably the blur that otherwise would have resulted from the motion parallax. But you can still tell that the craters near the horizon are kind of discolored.
Today I did the Wikipedia version with a refinement of the same technique. Only this time, I saved eight different versions of the IR/green/UV composite with the layers shifted to optimize the registration in different parts of the picture. Then, for each, I cut out a fuzzy selection around the optimization point, and put all of those into the image as colorization layers covering different parts of the moon.
Also, the rightmost part of the image had some data dropouts on every other scan line (I'm not sure about this, but I think these come from Cassini's data compression scheme when the planners underestimate the data allocation needed for the picture). In December I just hacked out that part. This time I got rid of the dropouts with a video deinterlacing filter, then used a piece of that version as a patch to cover over the damaged part of the main image.
The resulting image still has rainbow fringes in it if you look closely, but I'm getting better at this.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
no subject
Date: 2005-02-14 01:50 am (UTC)no subject
Date: 2005-02-14 05:28 am (UTC)no subject
Date: 2005-02-14 12:07 pm (UTC)no subject
Date: 2005-02-14 12:12 pm (UTC)no subject
Date: 2005-02-14 03:28 pm (UTC)no subject
Date: 2005-02-14 03:36 pm (UTC)no subject
Date: 2005-02-14 03:42 pm (UTC)