Noise Reduction in the Chrominance Layers

[twade]

Hi everybody,

Last night while processing an image, I noticed I was destroying a lot of star colors when implementing ACDNR on the luminance layers.  Obviously, what I thought was noise is actually defining the star colors.  What's the best way to prevent the loss in star colors when reducing the noise in the chrominance layers?  Should I be using the same luminance mask I use to protect features in the Luminance layer or do I have some parameter set too high?  Any thoughts or suggestions would be greatly appreciated.

Thanks,

Wade

[Carlos Milovic]

Hi Wade

As you said, a too much aggressive ACDNR application may destroy the colors of the stars. Mainly, there are two ways to protect them.

First, you may activate the luminance mask for the chrominance channels. Play with the midtones balance and the number of removed wavelets layers. The later will help you to avoid protecting hot pixels.

Second, decrease the star threshold, and do the same with the sides thresholds. Increase a little bit the Overdrive parameter (only for the chrominance, of course). A value of 0.1 may give good results.


Anyway, my philosophy with ACDNR is that losing a bit of color saturation in the stars is not that bad. Becouse I apply a multiscale threatment, and separate stars from the large scale features, I end with a "stars only" mask wich I use to apply a star shaping routine, wich thinners them and increases color saturation. So, the bottom line of this is that you may create a mask (the luminance itself may be a good choise for simplicity) and recuperate the saturation easily.

[Juan Conejero]

This is a good question. The fact that we can usually apply more noise reduction to the chrominance doesn't mean that we can be less careful with small-scale structures.

Usually, ACDNR's edge protection does a good job in protecting stars on the chrominance. However, there are difficult cases. Here's a small list of things that you can do to keep your stars healthy:

- Be careful with chrominance edge protection and star protection parameters.

- If necessary, use a luminance mask also for the chrominance.

- Consider using two instances of ACDNR: one for the luminance with its own mask, and another one for the chrominance with a different, more specific mask. For example, you can define a mask that just protects the stars for the chrominance.

- In really difficult cases, consider building a star mask with a wavelet-based technique, instead of the luminance-based mask that has been integrated in ACDNR. I think this is very unlikely to happen, though.

Hope this helps.

[twade]

Carlos and Juan,

I really appreciate your suggestions and fast responses.  I'll give them a try.  I'll also update my tutorial to include a warning to watch out that you don't destroy your star colors while working on the chrominance layer.

I end with a "stars only" mask wich I use to apply a star shaping routine, wich thinners them and increases color saturation. So, the bottom line of this is that you may create a mask (the luminance itself may be a good choise for simplicity) and recuperate the saturation easily.

I've always liked the star colors in your images.  They are excellently preserved.  Do you plan or have you created a tutorial on your star shaping routine?  Out of all the processing I do, protecting my star colors is my weakest link, and my goal is to get better.  Any help would be greatly appreciated.  For example, at what step do you create this mask?  If it's early on, how do you prevent it from "showing up" in your image with all the stretching and processing that needs to be done?

Thanks,

Wade

[Carlos Milovic]

Hi Wade

I've always liked the star colors in your images. They are excellently preserved. Do you plan or have you created a tutorial on your star shaping routine? Out of all the processing I do, protecting my star colors is my weakest link, and my goal is to get better. Any help would be greatly appreciated. For example, at what step do you create this mask? If it's early on, how do you prevent it from "showing up" in your image with all the stretching and processing that needs to be done?

Thank you very much  


Yes, I do plan a tutorial, but not too soon... My prioritys are now helping Juan to write de documentation, process a milky way panorama, and give my first steps with a DSLR.

Let's see anyway a quick description... I apply the star shaping routine at the very end of the image threatment. Of course, it is very important that the information must be protected/preserved throught all the steps. There are no miracles or magic, so the stars should not be saturated (white) discs.
Before the star shaping routine, I separate the stars and high frecuency features from the large scale ones. This is done with morphological filters and finally removing wavelet layers. So, I have two images, one containing the stars and details. After processing both images and merging, I discard the one havind the large scale features. To apply the star shaping routine I modify the high frecuency image to remove non stelar objects and the fainter stars (wich don't need star shaping). This is done with a very aggressive shadows clipping. After that, a morphological filter (minimum and then maximum) with a large circular kernel (15x15, for example) will remove the smaller stars that survived, and leave only those bigger than the kernel size. Then, remove the first 3 or 4 wavelet layers to smooth the profile of the remaining stars (the morphological filter turns them into flat discs, but don't worry, this is fixed with wavelets).
So, here we have at last the mask... if needed, fine tune it with histograms or the minimum/maximum filters.
The star shaping routine is based on several iterations of minimum filters, lets say 10 iterations, with an amount of 10%. The kernel should be small, no larger than a circular 5x5. After the minimum filter, adjust the luminance and CIE c channels with curves. Repeat both processes if needed (adjustint the parameters).

And that's all. Looks complex when reading it, but once you see it working, you'll get the feedback to understand what each step does, and why.