Bayer Drizzle and Outlier Rejection

Hi @minusman,

Excellent, that is good to know ;-). Yes, it's still better to supply enough frames when you apply drizzle/bayer drizzle integration and that you dither and have undersampled data to profit from it.

Regarding your question:

My question now is, I recently had an article about Drizzle in an Astro magazine.
read. It was pointed out that integration is only possible with average mode. Without outlier rejection, because the pixels of the finer grid are only alternately filled with content, which resembles the appearance of a disturbing pixel. These are sorted out with Median or Sigma Stacking. Would this also apply to Bayer Nieselregen?

First of all, drizzle and bayer drizzle are one and the same. The code in APP to perfrom drizzle or bayer drizzle is exactly the same piece of code 😉 The only difference is the input data, which in the case of bayer drizzle, is undebayered Bayer CFA data.

Indeed, if drizzle/bayer drizzle is chosen, the integration method (average or median) is forced to average. This is essential/required for drizzle/bayer drizzle integration. You can find this forcing actually in the console panel when drizzle integration starts.

Now with regard to outlier rejection, the article that you have read is rather incomplete, I would argue. It's true that due to incomplete drizzle per frame (which is perfectly normal) the pixels stacks will have zero values at those places of incomplete drizzle.

But it depends on the actual integration implementation, if this represents a problem to apply outlier rejection in integration. In APP, these zero values due to incomplete drizzle per frame, are disregarded in the pixel stack :-).

In other words, in APP, these zero pixel values are not considered in the integration, so the outlier rejection will still work properly, especially if you supply enough frames 😉

Kind regards,

Mabula

Hello Mabula, I did a test integration once. The pictures of h and x Persei I took last year with a Canon EF f/2.8 200mm lens and the ZWO ASI294 MC Pro.
There are 50 frames with an exposure time of 30 sec. pixel scale is 4.93"/pixel and all frames are sorted out without shifting.
The first integration is with pixel interpolation.

The second integration is with Bayer Drizzle, droplet size 1.0 and scale 1.0. In the details you can see that the SNR is higher and the noise lower and the resolution is a bit finer.

Many thanks for the improved Bayer Drizzle, with more frames I can definitely get something out of it with my lens / camera combination. 🤗 

I read the article about Drizzle again. It was on Autostakkert! 3, Registax, Fitswork and Deep Sky Stacker related. There that is probably problematic with the outlier rejection. Oh, and Dylan O'Donnell recently mentioned Astro Pixel Processor in his Star Stuff series regarding Drizzle.

Hi @minusman, can you please forward me the article and perhaps Dylan O'Donnell's mentioning of APP ?

One crucial aspect of drizzle is that integration is done with a weight map, to indicate how much % of a droplet lands on a pixel in the target pixel grid. These weights are used in APP's outlier rejection algorithms and the integration of the resulting pixel stacks.

I really think that if software has a problem with drizzle and outlier rejection, it's the actual implementation that rules the outcome, not the drizzle technique in principle 😉

It's important to realize that is you use, drizzle, especially Bayer drizzle, that the number of frames to integrate versus the ourlier rejection settings of kappa and iterations, is actually less than the real number of frames. This is due to the holes in the integration layers, due to the fact that not all pixels receive drizzle droplets.

So an integration of 100 frames, depending on the actual outlier rejection settings, can be regarded as integrating 50-75 frames with respect to the outlier rejection parameters.

Let me know if this is clear 😉

Cheers,

Mabula

Posted by: minusman

Hello Mabula, I did a test integration once. The pictures of h and x Persei I took last year with a Canon EF f/2.8 200mm lens and the ZWO ASI294 MC Pro.
There are 50 frames with an exposure time of 30 sec. pixel scale is 4.93"/pixel and all frames are sorted out without shifting.
The first integration is with pixel interpolation.

The second integration is with Bayer Drizzle, droplet size 1.0 and scale 1.0. In the details you can see that the SNR is higher and the noise lower and the resolution is a bit finer.

Many thanks for the improved BaExceyer Drizzle, with more frames I can definitely get something out of it with my lens / camera combination. 🤗 

Excellent @minusman. that looks very good ! Thanks for sharing 🙂

Mabula

Hello Mabula, I have sent the article to you as an e-mail. And here is the link to the video of Dylan O`Donnell with the topic Drizzle.

With kind regards. 🙂