Zion said:The 4-eyed stormies are a result of the rudimentary DVNR process, which I'm sure used some sort of temporal algorythm, but that's not really what I'm getting at here.
Yeah the DVNR included a lot of temporal smoothing, the 4 eyes occur because the position of the stormtroopers black eyes and black headband lined up in subsequent frames. Temporal filters always have problems with movement in the frame and should therefore be avoided (at least in my opinion).
The Photoshop filters that I've highlighted here seem more advanced than any simple temporal softening or averaging filter applied in Avisynth. The median filter doesn't just average all frames together, but seems to intelligently take the most consistent information and eliminate the anomalies like noise, burn marks, etc. (much like the TooT process).
but even using a median filter if you use subsequent frames from the same source you will get problems with motion, the only way to avoid it would be to mask out any areas that show movement (this can be done automatically in avisynth) then average the remaining areas of subsequent frames before re compositing the moving objects, but this would probably result in a noticeable difference in quality and any advantage is lost as soon as the camera moves.
I don't know if I've ever seen anything quite like it in a video filter. Most temporal noise filters tend to leave their own unwanted artifacts in the frame on anything with motion.
and your examples show these problems, i.e. Luke's 'trail' across the frame, the only reason it's minimal is because you averaged out a lot of frames and Luke doesn't stay in the same spot very long, temporal softening is only usually applied to 1-2 frames either side if you do this then you will see obvious blurring.
Sorry for ranting on about this but I really don't like temporal softening, it's a quick and dirty way of removing noise which removes detail on movement and destroys film grain.
Now onto something I do like, averaging of multiple captures to improve the quality of an analogue source.
What I'd love to see is how these Photoshop filters stack up (no pun intended ;)) against similar Avisynth functions when merging 5-10 captures of the same source.
The problem is this will not remove scratches and burn marks as they will be the same on every capture, theoretically a mean filter would be the best for removing analogue noise as it is a completely random effect, however the median filter would be advantageous in removing dot crawl, and the effects of laser rot which may not be the same on every capture.
We discussed different averaging methods using Avisynth in the V8 thread, and it isn't very easy to do a median of multiple frames with any in-built functions as it excludes them based on the difference between whole frames not on a pixel by pixel basis. However it would be possible to program a new filter that did a median for each pixel of the frame, or an even more detailed statistical analysis (assuming a large number of captures) taking into account the standard deviation etc. to deduce the center of the distribution of values for each pixel.
I would be interested in using this filter to average captures from different sources but without any temporal softening, what does it look like if you use the filter to combine your 'before' shots from the DC and the SC?