3D digital noise reduction (3D DNR): the trade-off between noise reduction strength and smearing
The difference between 2D and 3D noise reduction
2D noise reduction only performs spatial filtering within a single frame and smoothes adjacent pixels. It can suppress noise, But it will also erase the details altogether, making the picture look "blurry".
3D noise reduction adds a temporal domain dimension on this basis: comparing the same pixels in multiple consecutive frames, The static part is considered as image, and the changing part is considered as noise. So it preserves details much better than 2D noise reduction, This is also standard for current cameras.
Where does smear come from
The premise of time domain filtering is that "what is still is an image." When an object moves quickly, The algorithm will mix the content at the same position in adjacent frames, so the afterimage is dragged out behind the moving object—— The higher the noise reduction level, the more frames are involved in the comparison, and the more obvious the smear is.
Setting suggestions
| Scene | Noise reduction level | Reason |
|---|---|---|
| Static warehouse, perimeter | High (70~90) | The picture is basically static, no risk of smearing |
| Pedestrian passage | Medium (50～70) | Considering noise and walking afterimage |
| Roads and checkpoints | Low (30~50) or closed | Vehicles are fast and must avoid shadows |
| No fill light at night | Medium high | Noise is the main contradiction |
Practice reminder
- First solve the amount of light (target surface, aperture, fill light), and then talk about noise reduction. Noise reduction is just a "stop loss" method.
- Be sure to look at the still frames of playback when debugging. Real-time images can easily be misjudged due to differences in compression and display.
- When there is a big difference between day and night in the same scene, use the device's "day and night parameter separation" function to set them separately.