This filter reduces noise in video by quantizing pixel values, i.e. rounding them to discrete steps. Random, low-amplitude variations are removed because they fall below the quantization step and cannot be represented.
This method:
- Operates spatially (per frame)
- Does not analyze motion or neighboring frames
- Does not blur edges or average pixels over time
Noise is reduced by removing precision, not by detection or smoothing.
Parameters
Quantization (0–64)
Controls the quantization step size, i.e. how coarse the rounding of pixel values is.
Conceptually, pixel values are rounded to multiples derived from this parameter.
-
Lower values apply finer rounding
- Removes only very small variations
- Preserves gradients and texture
-
Higher values apply coarser rounding
- Removes more noise and fine detail
- Reduces tonal precision
- May introduce banding or posterization at high settings
This parameter defines which level of detail is discarded.
Strength (0–100%)
Controls how much of the quantized result is blended with the original image.
- 0%: original image (no effect)
- 100%: fully quantized image
Intermediate values interpolate between the two, allowing noise reduction while limiting visible artifacts.
Mathematically, the output is a linear interpolation between the original and quantized images.
Parameter interaction
The two controls serve different purposes:
- Quantization determines the rounding threshold and loss of precision
- Strength controls how strongly that rounding is applied to the final image
Using a higher Quantization value with a lower Strength often produces cleaner results than using low Quantization at high Strength.
Visual characteristics
Typical effects include:
- Reduction of fine grain and compression noise
- Better edge preservation compared to blur-based noise reduction
- Possible banding in smooth gradients if overused
Recommended use cases
- Archival, CCTV, or low-quality sources
- Footage affected by compression artifacts (like JPEG images)
- Situations where temporal noise reduction produces ghosting or smearing
Limitations
- Not motion-adaptive
- Removes detail indiscriminately below the quantization threshold
- High settings may degrade smooth gradients or skin tones
Summary
Quantization noise reduction removes noise by restricting the number of representable pixel values, with Quantization defining the precision limit and Strength controlling how much of the quantized image replaces the original.