Noise reduction option - specifically "smart" graininess removal

So, I phootgraph wildlife. I just started recently to produce clips also, but I’m not one to shy away from using high ISO, which by default produces graininess.
-Post-production for my photos, I use GIMP, and for reducing graininess, the best way I can, I actually use a filter that was not adopted as default for GIMP - called “wavelet denoise”. I have the source files if an adept person is willing and able to add such a thing to shotcut.

Basically, the filter uses RBG or YCbCr in preview window - but simple controls which include: ‘amount’ and ‘detail’ - which correspond to the number of cycles it runs (amount) and the pixel to pixel delta (detail). It seems average over the number of pixels in detail and to discard larger deltas and smooth only up to a calculated threshold.

The end result is amazing, since one can reduce background retaining for the most part almost the full sharpness of the image. Used over time, one can fine tune the two parameters to produce the right trade-off every time. -It helps having the preview frame to consider the impact each time the filter is used before applying it.

Adding this capability to Shotcut would make it an even better tool than it is now at least for photographers/clip-makers who like me often find themselves in less than optimal conditions for producing the source clips.

1 Like


Could you provide an assessment of the current noise reduction filters in Shotcut? In what way do they fall short?

I’m sure you must have your reasons but why not try to fix the problem at source and simply reduce the ISO?

I released a new filter just for you. I use it often, specifically to knock out high ISO. See this topic:

I tried to use the noise reduction functions in Shotcut. So far, the one that works best for me is: Reduce Noise: HQDN3D. It seems to be more refined that the other one, and have less of a softening effect on the sharpness.

If you shoot wildlife, you will find that the main activity is dawn-early morning, and dusk. Both do not offer you a lot of light. So reducing the ISO is not the answer.

The same goes for shooting in forested/ sheltered areas.

Moreover - the sensors are digital. They present you with an ISO range of 50-25,600 habitually. If you try and work on ISO 100, you’re working not in the linear response part of the sensor operational range, but towards its end. It would be like taking your car and driving it insistently close to its idle point or high revving it as a rule. Both produce deteriorated performance and in the long run don’t serve you well.
Other examples are operating an audio set at its minimal volume or maximal volume. This is not what you want as a rule, unless you’re “one of those guys”.

1 Like

Thanks a bunch Austin,
Since I now produce more and more clips, I will gladly use it and try and give you a feedback if you like, to finesse it (if I see anything to finesse/ fine-tune). Or will just reiterate my thanks if it’s all that I wished for.

OK - tried it vs. the "Reduce Noise: HQDN3D. My preferred preset is: 4-4-100 on soft. At least for now - It seems to be better than the aforementioned filter.

Thanks a bunch for this.

1 Like

Those numbers surprise me a little, but I also realize every image is different and has unique needs. Even though this is a wavelet denoiser, it is still possible for an image to get blocky or patchy in dark areas if the Decompose value is less than half of the maximum possible. But as always, whatever works! Thanks for giving it a try. Hope it helps.

Sure did Austin,

I can tell you that I use the GIMP wavelet denoise, using even odder numbers: I use it on a threshold of 0.5. This means to me that I’m trying to apply the wavelet math on a sub pixel level…(actually works for some reason). I think the same can be done to the Shotcut one you wrote, to make it able to impact the sharpness of the object in focal plane less, and smooth the background more…

Usually for me it is not about the art of the probable, but how can I use a filter to greater effect.

Thank you kindly for your help,