Log in

View Full Version : "Frequency" in temporal and spatial denoisers?


Fullmetal Encoder
1st April 2011, 20:29
Can anyone provide me with the precise meaning of "frequency" when discussed in relation to the way denoisers function? Does frequency refer to how often a pixel's chroma/luma values are changing from frame to frame? Or is it in relation to the absolute luma values with higher luma values being considered higher "frequency"? And what time frames are we talking about here? I would assume per millisecond as a per second frequency measurement would seem to be over an awfully long period for video processing.

Also, if we are dealing with a purely spatial process how can frequency be involved at all?

I am trying to obtain a better understanding of how dfttest, FFT3DFilter, TNLMeans etc., work.

J_Darnley
2nd April 2011, 09:58
If you perform a frequency transform, you will get frequency domain data. This is what a Fourier transform does (dft and fft). E.g. If pixels were changing from black to white every pixel across a frame, then that would be a high frequency. If here was a smooth change from black to white across the whole frame then that would be a low frequency.

Then frequency just relates to how quickly something changes. I don't think any of these tools account for framerate so I don't think they use an absolute measure of frequency (for temporal changes) but that is easy to test.

Gavino
2nd April 2011, 12:01
In everyday use, we are used to the word frequency relating to time (especially with audio frequencies), but more generally it refers to changes in one thing with respect to steps in another.

Frequency in sampling theory terms equates to changes per sample. Thus per frame for temporal frequency, and per pixel (horizontal or vertical) for spatial frequency.