19 June 2021: Our upload server https://upload.astropixelprocessor.com/ has been migrated successfully to our new office with higher upload and download speeds (nearly 10MByte/sec up/down ) ! We now have 1 general upload user called: upload with password: upload. The users upload1 - upload5 have been disabled.
31 May 2021: APP 1.083-beta2 has been released ! APP 1.083 stable will follow soon afterwards. It includes a completely new Star Reducer Tool, New File Saver Module, Improved Comet registration and much more, check the release notes here!
[Solved] Integrate : Pixel Interpolation
First of all, thx for this amazing software. Because of the weather not being nice lately, I'm just reprocessing some 2018 data.
I've tried many different settings especially in the integration part.
However, I'm not sure about the pixel interpolation choices and how it affects my images.
What are the main differences among all of them ? (Nearest Neighbour, Bilinear, Cubic B-Spline, Mitchell-Netravali, Catmull-Rom Spline, Lanczos 3, 4, 5)
Is there a rule of thumb which one to choose in certain situation ?
Hi Fred @fredmt,
Thank you for very much and welcome to the APP forum 😉
Basically, Lanczos-3 is the best in most cases and therefore default.
The lanczos algorithms are the best at preserving sharp details.
Algorithms like Cubic B-Spline, Mitchell-Netravali and Catmull-Rom Spline are good as well, but will have a slight blurring effect. These are nice when you want to downscale images. I would not recommend these for upscaling because you will lose sharpness.
Bilinear is very simple and will blur your data even more, and it will give ugly artefacts possibly on the stars in your images.
Nearest Neighbour is the worst when it comes to data resampling. It will not preserver detail and will give a lot of artefacts. It will however preserve noise the best of all algorithms.
Simply said, use lanczos-3 normally and when you are downscaling, you can try either Cubic B-Spline, Mitchell-Netravali and Catmull-Rom Spline. If you pixel peep, you will see slight differences.
Thank you very much for this very clear answer. Will try if needed.
I understand that the Lanczos order (Lanczos-3, 4 or 5) corresponds to the number of lobes kept in the interpolation. But, do you have a comparation with an image as example pointing out the differences? Because the sinus graphs I find are clear, but what is the result in an image?
You can simply check the difference by integrating only a couple of light frames, right?
Chances are that you will see hardly any difference 😉
Simply put, Lanczos-3 is almost perfect for all data, lanczos-4 and lanczos-5 will only slow down the data interpolation (larger filter kernel, so more calculations needed) and will add very little. I do know that if you have very large stars in your images (so very oversampled data), Lanczos-4 or Lanczos-5 can actually improve things a tiny bit so in that case you will visually see a difference in results when the registration parameters are applied.
Yeah, but I don't have any idea to where to pay attention to to notice the difference in the image. That's why my question if you had some examples, cause internet does not make me any wiser as well 😛
Thanks for the reply 🙂
Martin @martinsimmons, the differences are very! tiny, you will have a hard time spotting them I think. I never use Lanczos4 or higher 😉