DisplacementToMultiChannelDisplacement
Splits one selected image channel into low, medium and high frequency bands, stored in the red, green and blue output channels respectively. Connect the result to
MultiChannelDisplacementToDisplacement to adjust each band independently and combine them again.
Inputs
- Image
ImageThe image to split.
Output
ImageA three-channel floating-point image with the same resolution as the input. Each band is encoded asband * Scale + Midpoint. Values are not clamped to the range [0, 1].
Parameters
- Channel:
the input channel to split.
- Use Adaptive Blur:
scales both blur sigmas by
max(width, height) / 8192, using the input image resolution. Disabled by default. When disabled, blur values are used directly as sigmas in pixels.- Coarse Detail Blur:
the Gaussian sigma used to extract the low frequency band.
- Medium Detail Blur:
the Gaussian sigma used to separate the medium and high frequency bands.
- Midpoint:
the offset added to each band after scaling.
- Scale:
the multiplier applied to each band before adding Midpoint.
- Compute:
OpenGL or CPU. If OpenGL is not available, CPU is used.
Both blur parameters accept fractional values and must satisfy Coarse Detail Blur > Medium Detail Blur >= 0. They specify Gaussian standard deviations (sigmas), not kernel radii. A Medium Detail Blur of zero produces a zero high frequency band before encoding.
Adaptive blur
With Use Adaptive Blur enabled, the blur parameters describe an image whose longest side is 8192 pixels. The default values 16 and 4 were chosen for this reference resolution. The actual sigmas scale with the longest side of the input image, so resizing the same map while preserving its proportions normally does not require retuning the frequency separation.
Input resolution |
Coarse sigma (pixels) |
Medium sigma (pixels) |
|---|---|---|
2048 x 2048 |
4 |
1 |
4096 x 2048 |
8 |
2 |
8192 x 8192 |
16 |
4 |
16384 x 8192 |
32 |
8 |
The same resolution factor is used for both axes and both sigmas. Fractional sigmas are preserved. Downsampling can remove fine details, so the bands at different resolutions are not expected to match exactly.
Frequency bands
For the selected input channel I and the actual blur sigmas:
Low = GaussianBlur(I, coarseSigma)
LowPlusMedium = GaussianBlur(I, mediumSigma)
Medium = LowPlusMedium - Low
High = I - LowPlusMedium
Blur uses reflected image borders without repeating the edge pixel.
MultiChannelDisplacementToDisplacement reconstructs the selected input channel when its Scale and Midpoint match this node and all three Mix values are 1, subject to floating-point precision and any changes made to the band image.