IPC 3DNR Tuning for Non-Target Capture

Layer-based processing is supported, which means an image can be divided into foreground and background layers with corresponding temporal and spatial coefficients. This mainly involves temporal NR, with spatial NR as a supplement.

1. Set the three parameters of m1id0 and m2id0 to 0, 1, and 2, respectively. It is recommended that the pure spatial mode be used for level 0 and layer-based processing be used for level 1 and AdvMath be enabled for the foreground layer. Then, adjust the temporal NR parameters of the intermediate two levels (level 1 and level 2), including mXtfs, mXmath/mXmathd, mXmabw, and mXmate.
   • Layers 0 and 1 of level 1 may be respectively used as the foreground layer and the background layer. Adjust m1math/m1mathd of the background layer to suppress raindrop noise of the background image. Then adjust m1math/m1mathd of the foreground layer to control the noise of the foreground motion.
   • Adjust n1tfs at level 1 to suppress the overall noise level of the image, with the temporal NR effect as well as the motion smearing of the image carefully considered. Generally, you need to adjust tfs of the foreground layer to be smaller than that of the background layer to prevent the smearing of moving objects.
   • Adjust m2math and n2tfs at level 2 to control the quietness and noise amplitude of the overall noise. The value of m2math/m2mathd at level 2 is usually smaller than that of m1math/m1mathd at level 1 to prevent smearing.
   • In low light, to relieve the burden of raindrop noise suppression for math, you can adjust mabw. In extremely low light, you can adjust mate to suppress the overall image noise. In layer-based processing mode, the recommended value of mabw is 0 to 4 for the foreground layer. In AdvMath mode, the value of mabw for the foreground layer must be 4.
   • You are advised to use filter 6 as spatial filters at the intermediate two levels. Increasing the weight of filter 2 at level 1 improves the NR capability while compromising the edge-preserving smoothing. Increasing the weight of filter 4 at level 2 tends to strengthen edge preserving and reduce the noise in non-edge regions.
   • For layer-based processing at level 1, a certain ratio of spatial filtering results may be mixed into the foreground layer by increasing the value of n1tss of the foreground layer. This enhances the NR effect of moving objects.
2. Use the spatial filter 5 of level 0. Based on the current noise characteristics, adjust the strength of filters 2 and 4 and the blending weight of filters 0 to 4 to reduce the noise without losing the relationship area. You are advised to increase the NR weight for the small window and reduce the filtering strength.
3. Use pure spatial filter at the last level. m3math/m3mathd can be adjusted to separate the static region from the motion region and configure different spatial parameters for them. You are advised to use filter 5 in this case and use filter 6 to control the strength and range of filter 5. The weight of filter 4 needs to be increased. Adjust parameters of filter 4 to reduce the noise in flat regions and improve image clarity. The final result is output through combined effect of sth and sfn.
4. Adjust the parameters of the chrominance NR interface NRc based on proper tuning of luminance noise reduction.