ProPane Tweak: That Was the Tweak That Was

Directly Tweaking Image Pixels

We will start with out usual demo image:

image = Rfits_read_image(system.file('extdata/VIKING/mystery_VIKING_Z.fits',
  package="ProFound"))

plot(image)

Below we shift the image by x +2, y +2 and rotate it by 5 degrees (clockwise):

image_tran = image
image_tran$imDat = propaneTran(image_tran$imDat, 2, 2, 5)

And we can check this:

plot(image_tran)

Now we tweak the solution (this will take about 20 seconds):

tweak = propaneTweak(image, image_tran, delta_max=c(5,10), shift_int=FALSE)
#> Trimming comparison region and auto scaling
#> Optimising tweak
#> Creating output image_post_fix

Our tweaked image is pretty close to correcting for our original transform:

best = tweak$optim_out$par
print(best)
#> [1] -1.898162 -1.431492 -4.513166

We can use the output to manipulate the pixels:

image_fix = image
image_fix$imDat = propaneTran(image_tran$imDat, best[1], best[2], best[3])

plot(image_fix)

We can also modify the WCS rather than the image pixels:

image_fix2 = propaneWCSmod(image_tran, best[1], best[2], best[3]) #correct WCS

plot(image_fix2)

In the last example note how the WCS grid follows the original sources correctly.

To make all of the above easier we can request propaneTweak to directly modify our WCS passed in (which is the default):

plot(tweak$image_post_fix)

Directly Tweaking Image WCS

Now we will look at a case where the pixels are matched but the WCS is not:

image_tran2 = propaneWCSmod(image, 2, 2, 5)

plot(image_tran2)

Note we set WCS_match to FALSE since it is the WCS itself that is wrong.

The tweak of the WCS directly will takes longer than before (about 100 seconds):

tweak2 = propaneTweak(image, image_tran2, delta_max=c(5,10), shift_int=FALSE,
  WCS_match=FALSE)
#> Trimming comparison region and auto scaling
#> Optimising tweak
#> Creating output image_post_fix

The answer almost perfectly corrects are adjustment to the WCS:

print(tweak2$optim_out$par)
#> [1] -2.000074 -2.000107 -5.000139

And the image and WCS looks very nice now:

plot(tweak2$image_post_fix)

What Sort of Tweak

Most of the time users will want to use the WCS_match = FALSE form of tweaking when trying to improve the registration of slightly offset frames for stacking. In this situation they will need to choose a reference frame to tweak all subsequent frames to, or be careful to tweak frames in the correct order so that a larger mosaic is all correctly registered.

Given the computational cost, it is often a good idea to determine tweaking based on a nice subset region of the image (ideally only a few hundred to a thousand pixels in dimension). Tweaking a very large image (8k x 8k) will take quite a long time (probably hours), but 1k x 1k should be around 10-15 minutes.