y3llow, why would you need to do temporal NR on a RAW sequence? There is no inter-frame referal of data?
By "no inter-frame referral of data' you mean because it's not "video"? Every frame refers to each other as it's in motion, we view it in motion? :-)
By temporal NR I mean that the NR tool does motion analysis, creates vectors and the noise reduction algo's check frames backwards and forwards, maybe just a couple each way, maybe 10 or 50 each way depending on processing power.
The algos attempt to establish what over time it assumes to be detail, noise, high frequency jitter and shimmer and sets about reducing it based on users set parameters and the movement or lack of movement in the noise.
Noise reduction in singular image processing is a subjective and visual process, we adjust the sliders, we decide on the parameters for a single image, until we're happy with it, at the point that a balance is struck between noise and detail.
But at 24fps we're going to need something a bit more than profiling the camera and creating a dark frame.
I think noise reduction is best done in a number of small increments targeting specific problems whether it be fixed noise, shimmer, flicker whatever. And then the adding of controlled levels of noise back in to avoid banding.
Really don't see how anything other than a temporal approach would be beneficial.
