Fractional Pixel Image Translation

areeves

Member
Mar 11, 2015
9
1
I'm trying to create a Neowise comet mosaic and have a series of subs that crop out the nucleus;. There's enough overlap with the frames including the nucleus that I could StarAlign all of them using the same reference frame. However because I didn't shoot a frame of the nucleus *after* shooting the tail frames, I cannot use CometAlignment to offset the 'tail-only' frames. This is because Comet Alignment only uses the comet nucleus location in the first and last frames to make its offset calculation (it could be more flexible this way*).

I want to translate these 'tail only' frames so they can be part of a final comet-aligned mosaic. One way would be to

1) Use Crop to expand the frame of each subframe large enough to include the calculated location of the comet nucleus and
2) Place a copy of the comet nucleus in it's calculated location the last 'tail only' frame
3) Use CometAlignment to offset all the frames to align the nucleus.

If there are better ways - enlighten me!

Regardless, I wonder:

I can perform INTEGER-pixel translations using PixelMath & Dynamic Crop. But is there any way to perform fractional pixel translations without coding resampling algorithms in PixelMath or resorting to the Alignment processes? It seems that Dynamic Crop should allow fractional pixel anchor point changes, as it is prepared to resample images.

Thanks for your help,
Arlo

*Feature requests for CometAlignment:
1) Allow extrapolation as well as interpolation of comet nucleus position. That alone would address this issue.
2) Allow the user to locate the comet nucleus on more than one frame, performing a least squares fit to best match its movement through the subs where it's visible and, with the new exrtrapolation, to subs of that don't include the tail. This would improve comet nucleus registration since deep comet images may overexpose the nucleus, limiting the accuracy with which CometAlignment can 'automagically' locate the center of the nucleus.
 

dld

Well-known member
Sep 24, 2017
161
14
Hi Arlo,

What if you use the ChannelMatch process (o_O) with equal offsets for each color channel? The offsets provided by the CometAlignment process may be useful.

BTW a "Frame Interpolation" process would be useful to translate images in order to reduce/eliminate star gaps caused by missing frames when performing Comet Integration :)!
 

areeves

Member
Mar 11, 2015
9
1
Thanks dld, that's a good workaround. I also discovered that if there's a *tiny* rotation or scale (0.00001 or 1.0000001) then it will trigger a resample in the Dynamic Crop process (Ok when relocating a copy of a 40 pixel square comet nucleus). Fractional-pixel anchor point offsets don't trigger resampling, but appear to round to the nearest integer.
Arlo
 

areeves

Member
Mar 11, 2015
9
1
I've since realized that CometAlignment does not require visible comet nucleus, just the offsets which can be typed directly into the process panel's Parameter's pane simplifying the process of comet-aligning frames with no nucleus, especially if the reframing involves translation and no rotation.
Arlo
 
  • Like
Reactions: dld