Noise Reduction using MMT
Noise Reduction (NR) is part of most processing workflows for astro-images. There are a lot of different tools designed for NR. The challenge with all of them is to find a balance between smoothing out noise in the background without loosing detail and sharpness in the object. Secondly, we don’t want to introduce artefacts such as worm-like structures or weird spots in the background. For a while I have been using MureDenoise, which works pretty well. But dialing in the right settings can be very time-consuming.
Recently, Benjamin Csizi shared on one of his Astrobin images some details on the NR he used. It was essentially the technique originally described by Jon Rista. It works on both linear and non-linear images. So far, the outcomes have been very consistent and NR is applied in a very natural way, without introducing artefacts. Let’s walk through the various steps as follows.
Make a Mask by cloning the image you want to edit.
Stretch this mask, for example using Process>IntensityTransformations>MaskedStretch (only if working with linear image).
Adjust this mask with Process>IntensityTransformations>HistogramTransformation and ensure that the luminance peak is at around 50% (for linear images) or 75% (for non-linear images).
Reduce the contrast of your mask using Process>IntensityTransformations>CurvesTransformation by pulling down the highlights.
Stretch the Mask to about 50% (linear) of 75% (non-linear). Then reduce contrast by pulling down the highlights.
Apply the Mask over your image and invert it.
Now apply NR using the tool Process>NoiseReduction>MultiscaleMedianTransform. Choose 8 layers and set the parameters as in the below screenshot.
Look at the result and determine if you like the outcome.
Tinker with the masks if you want to adjust the results
The amount of NR applied can be influenced by the stretch in the mask. Be aware that the mask is inverted, so the lighter the mask, the less NR. If you make the mask darker, more NR will be applied. If you want to fine-tune the NR in brighter parts of the image, such as your object, you can apply that selectivity by using the curves adjustments.
In my last three images, NR was applied in the way described here and proofed very reproducible and effective. If you save the process-icon with the appropriate MMT settings, you can just load the tool with all settings each time you need it, without having to fill in parameters for all eight layers.
If you want, you can follow-up with other NR techniques, such as TGVDenoise for high frequency noise. But this may not always be necessary.
For the final effect, see the two images below. On the left the image before NR was applied and on the right the result after NR was applied as described above.
