L
Nik Szymanek
Imaging Masterclass: From pixels to
gigapixels (part 2 )
ast issue we used Topaz Labs’ Gigapixel AI to resample some daytime DSLR images and then
assessed the end result after enlargement. e images passed muster with ying colours,
outperforming the resampling tools in both Photoshop and Affinity Photo. is issue, we’ll examine its
performance on deep-sky astro-images, and things are going to get really interesting in a way that I
certainly wasn’t expecting.
My curiosity about Gigapixel AI was piqued by the fact that when I’m imaging in my back-garden
observatory I tend to bin my CCD camera into a 2 × 2 con guration. is works well since it makes
the camera more sensitive and the pixel sampling rate matches the typical seeing conditions. e
end result, though, is quite a small le, around 1,663 × 1,252, which just lls my PC screen when
shown at 100 per cent. And, of course, resampling is required for any poster or large-print
production. Having good software to x this is important, so let’s see how well Gigapixel AI performs
on the deep sky.
I started by calling up a 2 × 2 binned LRGB image of the beautiful galaxy Messier 81 in Ursa Major
(Step 1). A quick application of Gigapixel AI to enlarge it by 2× produced an intriguing result. After
manipulating the ‘Suppress Noise’ and ‘Remove Blur’ sliders I was very surprised to see that not only
had it performed very well in terms of resampling, but it had also sharpened the image, better
resolving the detail in the spiral arms and tightening up the star images and diffraction spikes.
Further, the image was smoothed, with no loss of resolution (see the right pane in Step 1). is
really piqued my interest, so I opened another image taken a few years ago from the AstroCamp
remote observatory in Spain. e galaxy featured in that image is Messier 82, which is also in Ursa
Major, and this image in particular is an extremely deep LRGB image combined with hydrogen-
alpha data to capture the bipolar outburst from the galaxy’s core. I applied 2× resampling and then
once again adjusted the two sliders and was amazed at the level of detail that popped out of the
image (Step 2). is was particularly evident in the dark, dusty tendrils emanating along the plane
of the spiral arms and in the out ow material. As before, star images improved, as did the diffraction
spikes. And, yet again, the image was smoothed remarkably well.
I certainly had not expected these enhancements from a resampling software package. When I
compared the before-and-after views shown in Steps 1 and 2, I was once again reminded of
deconvolution software, whose raison d’être is to eke out resolution from blurry images. Gigapixel AI
does that almost as an afterthought and with no ugly ‘ringing’ around the stars.
For the next test, I wanted to try resampling an image that contained lots of stars as well as
nebulosity. For this, I used an image of the famous Eta Carinae Nebula that was taken with the
Telescope Live remote-imaging platform in Australia. Using a Takahashi FSQ-106 premium
refractor coupled with an FLI PL16803 large-format CCD camera and Astrodon narrowband lters,
the entire nebula was visible in a single frame. Step 3 shows the image in the left-hand pane. I
usedGigapixel AI to resample the image by 2× and then cropped it to home-in on the lovely emission
nebula NGC 3324, which is seen at the top left. e centre pane shows the enlarged original crop
zoomed in to 200 per cent to match the scale of the 2× resampled version, shown in the right-hand
pane. e original image is quite pixelated, but the resampled version shows great stars and nebulae
with just a small amount of smoothing.
Next, I opened a Milky Way image taken under La Palma’s dark sky with a Canon 5D MkII DSLR
and made up of ten 300-second exposures stacked together. Once again, I applied a 2× resample and
adjusted the sliders to get the result shown in Step 4. To obtain this image, I’d pre-focused the lens
in twilight using the camera’s auto-focus capability and taped the lens barrel in place. is method
delivers the sharpest star images. On the left is the original image, showing the famous ‘Pipe
Nebula’ (also known as Barnard 59, 65–67 and 78) and on the right is the resampled and slider-
tweaked preview. It’s quite similar (no mean feat for a 2× resample) and has slightly sharper stars.
Clicking on ‘Save’ processed the image. I should mention that when using this program it doesn’t
save the image as a new state (like the other Topaz Labs programs) but creates a new layer. You can
hide the new layer to see the preprocessed version and compare the two. Intriguingly, both states are
shown as the same size. I’m not sure how that’s done, but it makes the comparison very easy.
If I’m happy with the result, then I atten the image into a single layer, hiding the original image.
It’s then a simple matter to just save the new image (Step 5).