How to reduce noise in a high ISO photo
Paint by Zaweba cleans high ISO noise with Profiled Noise Reduction, which learns your camera's noise, plus Non-Local Means and Wavelet Denoise.
High ISO noise is best removed by a tool that knows what the noise looks like. Paint by Zaweba's Profiled Noise Reduction measures a noise profile for each colour from flat areas of the photo, evens out the noise and then removes it with wavelet shrinkage, keeping detail that a plain blur would wipe out. The Noise filters also offer Non-Local Means and Wavelet Denoise, two other well proven methods.
Shooting indoors, at night or at a fast shutter speed pushes the ISO up, and the picture fills with speckles: grainy brightness noise and blotchy colour noise. A good denoise removes most of it while keeping edges, texture and fine detail.
What you need
The photo, ideally the original file rather than a compressed copy, and Paint by Zaweba on Windows. If you shot RAW, the RAW Developer has its own noise reduction in its detail settings, and it's often best to start there, since it works before the picture is turned into finished colours. Everything runs offline.
How Profiled Noise Reduction works
Noise isn't the same everywhere. It's usually stronger in the shadows than in bright areas, and different in each colour channel. Profiled Noise Reduction measures this from flat areas of the photo, such as a wall or a clear sky, where any variation must be noise rather than detail. You can press the button to measure, or let it measure automatically.
It then applies a mathematical step called a generalised Anscombe transform, which makes the noise even across dark and bright areas, so it can be removed with the same strength everywhere. Finally, wavelet shrinkage takes out the noise at each scale of detail. The result is usually cleaner and more natural than a single blur.
Step by step
Open the photo and choose Profiled Noise Reduction from the Noise filters. Let it measure the noise, with the button or automatically. Look at the preview at 100 percent on an area with both flat tones and fine detail, such as a face against a wall.
Raise the strength until the speckles go, then check the fine detail. Hair, fabric and text are where over smoothing shows first. Back off if they start to look painted.
Save the profile as a preset. Profiles belong to a camera and ISO, so a preset for, say, your camera at ISO 3200 can be reused on every photo taken at that setting, without measuring again.
Non-Local Means and Wavelet Denoise
Non-Local Means compares small patches across the photo and averages similar ones, which is very good at keeping repeated textures and edges. Wavelet Denoise works scale by scale, treating fine grain and larger blotches separately, which helps when one kind of noise dominates. Try them when the profiled method isn't quite right for a picture.
Getting a good result
Denoise first, then sharpen. Sharpening noise makes it worse, so clean the photo before any sharpening step, and sharpen through a mask so flat areas stay smooth.
Don't aim for zero noise. A little fine grain looks more natural than plastic smoothness, especially on skin. Many photographers deliberately leave a trace, or even add a fine grain back after denoising.
Treat colour noise more firmly than brightness noise. Colour blotches look ugly at any level, while fine brightness grain can pass for film.
Work on a copy of the layer, so you can mask the result. Shadows may need more noise reduction than the subject, and a mask lets you apply it unevenly.
Avoiding noise in the first place
Exposing well, even at high ISO, gives less noise than underexposing and brightening later. When you can, use more light rather than more ISO. When you can't, Paint by Zaweba's denoise tools make high ISO photos far more usable.
