Processing aggressively against noise and star trails

The thing is that I revisited some old pictures, taken with my Olympus DLSR from Grand Canyon NP in Arizona. Under the darkest and most wonderfull sky I’ve ever seen, I tried to grab a bit of the Milky Way show. I got 7 frames, at ISO 400, and 30-second long each.

When I tried to process them. I discover that they were very noisy, and, even worse, the stars had left long trails, spoiling the picture, so I forgot about it. Until three years later, when I tried to apply aggressive processing techniques to them.

This is the result. Of course, it does not deserve a place in the hall-of-fame of the Milky Way pictures, but anyway it is far more beautifull than I previosly thought.

Via Làctia des del Gran Canó del Colorado

Although processing DLSR images is not my speciallity, I decided to write a bit about this, as some of the techniques are equally useful for CCD pictures.

After stacking the 7 frames, I inspected the result with PixInsight. The noise was remarkable, and I quickly decided that my first processing option, deconvolution, would be of little use with such a noisy image. As you know, deconvolution makes sense if you work with a linear image, that is, beore you apply any stretching to it. But the noise was everywhere, and even using masks to protect some parts, deconvolution was to yield artifacts.

So, I first stretched the image, a bit aggressively, to get rid off so much noise as possible. I would use deconvolution after this stretching, as a necessary exception to the rule.

This is a small fragment of the noisy image, where you can also see the star trails: noise and star trails 1

Now, the goal was to use deconvolution to reduce the trails, and get a as much rounded shape as possible. A overwhelming task, isn’t it?

First step is to generate a PSF for the deconvolution. This PSF (Point Spread Function) is a simulation of our stars shape, and PixInsight includes a function to do so. You must select some of the stars, and the function generates an artificial “star” (the PSF). This is my PSF:

PSF

As you see, it has captured more or less the nasty shape of my stars! With this PSF, deconvolution will try to get them back to a rounded shape.

But we cannot apply deconvolution yet. Although we have stretched the image, and got rid of a lot of noise, we must protect the background, as using deconvolution on it would create a mess. For this protection, we’ll use a star mask:

star_mask1

To create it, we must assure that the noise doesn’t get into the mask, fooling the function and creating false star masks. To do this, we can use the “noise threshold” in the mask generation function in PixInsight. This parameter works as a gate, allowing more or less signal to pass to the mask generation routine. Increasing it, we’re blocking noise to go forward. Of course, we also block little stars, but this won’t be a major problem for us, as we’re trying to get a wide-field, pretty image, so the smallest details will be anyway lost in the view.

With the mask applied, we go with deconvolution, using our PSF. I used 80 iterations. We can see the outcome in the same uggly fragment that we saw before:

noise and star trails 2

See the improvement! As usual, deconvolution has created the well-known artifact around stars: those dark rings. In a detailed CCD image, this would for sure be a problem. But, as I said, we’re interested in a large format picture, using all the available DLSR field, so those artifacts won’t be seen and we can get by with them. In other words, tha overall balance of using deconvolution is positive, regardless of the artifacts. See that the background noise has not worsened, as we have protected it with the mask.

Final adjustements have to do with aesthetic. Curves and color intensity adjustements, but only applied to the nebulosity. To do that, we create another mask, this time working with the Luminance component, and aggressively stretching it to get a shape of the protected area:

L mask

Using this mask, we can push color intensity inside the Milky Way. And using this mask, but inversed, we can adjust, with curves, the brightness of the background, lowering it as much as we can to hide noise.

Finally, I used the noise reduction function in PixInsight, both for the remaining background and for the most part of the Milky Way. Again, I’d have never done so in a regular, CCD image, as the detail would have got lost in the process. But this is a small problem for us in this picture, and we want a pretty looking, wide-field frame.

 

Leave a Reply

Your email address will not be published. Required fields are marked *