Noise Reduction in TMPEG

[perdomot]

Has anyone experimented with the NR settings in TMPEG? The only settings I've seen are 100-2-100 and they seem to work fine but the quadruple the encoding time of a file. I want to try and minimize any blocks or artifacts in my final mpegs. Thanks.

[snowmoon]

I would look into de-noiseing in AVIsynth. It would be impossible to even scratch the surface of what it can do, but I have switched to doing 100% of my alterations through AVIsynth and it's a wonderfully fast and elegent system. Many encoders swear by it.

[txpharoah]

If 80-1-80 won't fix your problems, your video is hosed. Also realize you won't see as much detail on your tv, so the effect will be the tv "cleans" it even more.

I normally use 20-1-20, 40-1-40, 60-1-60 and 80-1-80. Upping the middle number does very little (the tv will make it negligible).

[mrtristan]

I've been experimenting with them for a while. I've been using 100-4-100 for the last few captures that I've done. I also use the sharpen edge filter to resharpen the lines in the show, since the maximum noise reduction settings tend to blur the lines a little.

I've also used 40-2-40 often.

These are settings for anime captures, so they probably won't apply to real life video. I tried the maximum settings for Men in Black II and the video just got blurry. Anime doesn't have the same complex shading that real video has, so I'm assuming that's why the maximum settings don't destroy the integrity of the video.

I also tried experimenting with various Virtualdub filters for about two weeks, since most of them are faster than Tmpgenc. I couldn't find a combination that I was satisfied with.

[Thought Copper]

Has anyone experimented with the NR settings in TMPEG? The only settings I've seen are 100-2-100 and they seem to work fine but the quadruple the encoding time of a file. I want to try and minimize any blocks or artifacts in my final mpegs. Thanks.

How does your source AVI look like? Is it noisy with RF static? Or does it have already some macroblocks that you want to remove?

Tmpgenc's NR is good for reducing static but not the best for removing blocks that are already present in the AVI file. Too much noise reduction in Tmpgenc can remove too much color gradations making the video look overly smooth and CGI-ish.

VirtualDub has a SmartSmootherHQ filter that is made for reducing the appearance of blocks.

Again, if you post a picture of your source, I can better recommend filters to try out.

[IJM]

Hi,

I'm no expert, but having now had some limited experience with vdub noise reduction filters, I had another look at TMPG's noise filter and realised how it seems to work.

You have three sliders: "Still", "Range" and "Time Axis".

Time Axis seems to be a temporal filter. It compares each pixel with it's equivalent in adjecent frames. If its within the threshold set by the Time Axis slider it will try to smooth the pixel with those in adjecent frames. Range seems to have no effect on the Time Axis setting. Having said this, high settings don't seem to cause the ghosting you would normally expect from a temporal filter, so maybe there's something else going on here.

Still and Range seem to work together. These seem to be the same as the Threshold and Diameter settings in most 2d cleaners. They work on each frame in turn with no influence from other frames. The range controls the number of pixels to be examined in order to arrive at a value for the current pixel. Range 1 makes it examine the pixels immediately surrounding the pixel in question (ie. a 3x3 grid) in order to calculate the value of the centre pixel. Range 2 looks at a 5x5 grid, Range 3 a 7x7 grid and Range 4 a 9x9 grid. Pixels close enough to the value of the centre pixel as dictated by the threshold indicated by the "Still" slider are averaged out and used to arrive at a value for the centre pixel.

For better explanations of how "temporal" and "2d" noise filters work, examine the descriptions of various Avisynth and vdub filters.

Temporal filters can be quite subtle and can seem to stand relatively high settings without too many side-effects. However, 2d filters (as controlled by the Still and Range sliders) are enormously powerful and it is all too easy to loose subtle detail or just simply blur everything.

Run time is mainly dictated by the Range slider. As this value increases, the run time increases exponentially. This is because you are asking it to examine exponentially more pixels for each pixel in the movie.

One thing I have found is that while TMPG's noise filter is quite good, it seems to be much less efficient than some of vdub's filters. Furthermore, it seems that the best practise for subtle noise reduction without losing too much detail is to make several passes with noise filters at low settings rather than fewer passes at higher settings. For these reasons, I think I'll stick with vdub filters rather than using TMPG exclusively.

Regards.

Ian.

[NightWing]

I like it. Esp working on old VHS tapes. It really add to the processing time but does exactly what I want. Also good with recorded TV shows that are not that good in video quality. That plus a bit of sharpness and a color boost helps to clear up poor tapes.

[Ejoc]

AVISynth, I swear by it!

[energy80s]

AVISynth, I swear by it!

I swear at it!

[SatStorm]

The combo of the temporal cleaner and dynamic noise reduction on Virtualdub, blows up tmpgeng's built in noise filter IMHO. Expecially for VHS sources. IMHO always.

[C53248]

My own experiance with TMPGEnc's noise reduction has been better than with Virtualdub's various filters. The Virtualdub filters smeared and ghosted, including the NR during capture, the temporal smoother, and Klaus Post's Smart Smoother. Maybe some people found a better collection of filters and settings than I was able to. But TMPGEnc's filter never does that and has worked great for me.

For animation, I generally use 100 1 100 or 100 2 100. Animation usually doesn't have such faint detail that gets washed away by that. By the time it goes through the TV it looks perfectly clean and still retains all of the original artwork. Increasing the range from 1 to 2 seems to be the only setting here that affects the encoding time.

For live-action video, those settings will make people's hair look like a ball of clay, and everybody has skin as smooth as a plastic doll. After a lot of initial experimenting, I used 0 1 100 when I made VCD's. That is, 0% still picture and 100% time axis. That drastically reduced the random frame-to-frame color variation from NTSC/VHS with minimal effect on spatial detail. For DVD, I can use over 4000 kbps, enough to describe the noise on a good tape, so I don't use any filtering, and the result looks as good the original.

The TV does filter out a lot by itself, and noise that looks gross on the computer monitor is often nearly invisible on the TV.

For reducing MPEG blockiness, "Soften block noise" on TMPGEnc's Quantize matrix tab, set to 35 35 helps a lot without doing noticable damage. It doesn't seem to add any encode time. Blockiness can be caused by having too much noise to encode, though, so noise reduction of the source can also reduce it.

[vhelp]

.
.
I've found that, if your source is clean (ie, satellite) and you are using an
analog capture card, (and thanks to the noise that these cards put out)
you can do very well w/ TMPG's default Noise Reduction filter. The only
problem though.. its much slower. But, it does put out pretty good results,
if not better. It all depends upon your source (ie above) and the station's
quality etc. and your capture device. Most, if not ALL analog capture cards
produce Line Noise. External video capture devices like the ADVC-100
does not produce any Noise. So, any final filtering would only be dependant
upon the source quality from the station airing the source.

-vhelp