This is yet another filter which cleans up the picture, averaging out visual noise between frames.
The filter is designed to cope with the oddities of broadcast TV. It automatically fine tunes itself for good or poor quality signal and reception, and can even deal reasonably with some kinds of interference. It’s spiffy!
The Adaptive Noise filter examines the image, infers the amount of noise in the picture, and uses that estimate to determine how to cancel the noise. To differentiate between noise and motion, the filter makes use of correlations within and between fields as well as local color differences. As with the Gradual Noise filter, the degree of certainty about motion and noise is used to decide what mix of the previous and current colors to show.
The Settings
Noise reduction
Sometimes the filter doesn’t know whether change in the picture comes from noise or motion. When that happens, there’s a tradeoff — Guess motion, and the picture will be noisier if you’re wrong. Guess noise, and a mistake will blur the movement.
The noise reduction setting lets you tell the filter how to make this tradeoff. Set it low, and the computer will play it safe, avoiding any parts of the image which might be in motion. Set it high, and noise will be greatly reduced — at the cost of blurring and loss of detail in slowly moving areas. This tradeoff is affected by the amount of noise — The more noise there is in the image, the more blurring will be necessary in order to reduce it.
There’s no perfect way to choose this — Just move the slider until it looks as nice as possible. Comedies and dramas are probably the best material for picking settings, since facial close-ups make any blurring really easy to see. (That’s because we’re especially good at seeing distortion in faces.)
There are two kinds of video which deserve some special consideration: cartoons and sports. With cartoons, you can use a much higher noise reduction, since animation tends to have less fine detail. And with sports (or any other show with constant motion), there isn't anything to gain from a noise filter, so you can just turn it off. On the other hand, changing the settings all the time is a pain, so you might just want to choose a value and forget about it.
Before you fine tune this setting, make sure that the filter has adapted to the current signal. To do so, make use of the lock dot, described below.
Stability
To preserve a small amount of noise throughout the picture, move this setting down. To get a really solid looking picture, move it up. Don’t set it too high, though, or you’ll get posterization.
Regardless of your stability setting, the filter will always preserve a certain amount of variation in order to maintain the color depth of the original image.
Lock dot
It’s... a small green dot!
Check this setting, and a tiny green dot will appear toward the upper left of the screen when the filter’s estimate of the noise is confirmed by the current picture. That’s an indication that the filter has settled on a noise value. In general, the dot will turn off when all of the picture is in rapid motion — When there’s too much motion, the filter tends not to believe the current estimate. In that case, the filter makes do by weighting previous good values.
If the picture is constantly in motion, it’ll take a while before the filter can figure out the noise level. Interference also tends to make noise estimation take longer — The estimation time depends on the kind and amount of the interference.
If all else fails, tune to the signal with the interference, then quit and restart DScaler. The Adaptive Noise filter is more tolerant of interference right after it starts up, before it has seen a clean signal from another source. But with certain kinds of heavy interference, the filter will never be able to figure out the noise level. In that case, the Temporal Noise filter is probably your best bet.
Fast memory access
You’ll probably want to keep this setting turned on. With it enabled, the filter will run up to twice as fast.
However, there are a few motherboards — mostly made by VIA — for which Fast memory access will cause problems. If you see fleeting horizontal lines in your picture, try turning this off in any filters which offer it. If that fixes the lines, you’ll know to keep this setting disabled.
What’s the best way to get rid of noise?
Make sure you have a good signal. Noise can come from cable jumbles, poor connections, poor power and grounding, poorly designed video input cards, electrical gadgets (anything from a dimmer switch to various computer components) or from a bad video source. These issues are all beyond the scope of this help file. I’d suggest a look at the AV Science Forum ’s Home Theater Computers FAQ and board, where these topics are discussed at length.
Should I use a noise filter?
It depends what you’re watching. Sports and nature shows in general — really anything with lots of moving low contrast texture — are not handled well by the noise filters. That’s because those textures look a lot like noise.
The Adaptive filter does the best with difficult material, disabling itself where it detects motion. It causes surprisingly few problems with field sports so long as the background noise isn’t too bad. But it isn’t perfect — Road races are especially prone to blurring. When that happens, it’s best to turn the filter off.
Otherwise, you generally should have a noise filter running.
How much noise reduction should I use?
Let your eyes be the judge. I like to keep the settings pretty low — Blurred faces look worse than a little static.
Surprisingly, small amounts of color variation can sometimes improve an image. By switching back and forth between colors, the picture is able to give the impression of a color somewhere in between them. Also, noise can break up artifactual patterns in the picture, making it easier for you to ignore the errors.
With that in mind, the Adaptive and Gradual Noise filters and the Greedy (High Motion) median filter all try to preserve a small amount of color variation. As a result, they will never give you a completely stable picture.
Which noise filter should I use?
If your computer is older than a Pentium 3, Athlon, or some recent Celerons, the Adaptive noise filter won’t show up in the menu — because it can’t run. In that case, you should use either the Gradual Noise or Temporal Noise filters, which work fine on older computers. If you can run any filter, Temporal Noise and Gradual Noise are roughly the same speed at about 100 MHz. The Adaptive Noise filter is much slower, using about 200 MHz.
The Temporal Noise filter can give you a very solid picture. The Gradual Noise filter doesn’t look as solid and has trouble at high noise reduction values, but has less speckling and posterization and better color depth. The Adaptive Noise filter has a solid picture (if you want it), avoids speckling, color artifacts and posterization, and corrects for varying levels of noise. (Yes, it’s my favorite. But I wrote it, so I’m biased.) With a poor signal, the Adaptive filter can take a while to stabilize on a good picture, which can be annoying.
A fourth choice is an option hidden in the Greedy (High Motion) deinterlacing method. One of its abilities is a median filter, which works well for mild noise reduction with minimal harm to the picture. It delays the picture a little, making use of colors in upcoming fields in order to figure out how to reduce noise. The median filter is very slow (taking about 700 MHz), but you get excellent deinterlacing as well as the noise reduction.
In general, the more noise filters you run at once, the worse the picture will get. In particular, don’t run the Temporal Noise filter at the same time as the Gradual Noise or Adaptive Noise filters. It looks awful.
Can any of DScaler’s other settings help reduce noise?
In the Advanced Video Flags dialog, there are two (well, three) options which can help. “Horizontal Filter” will smooth the picture, cutting down on noise. You can also reduce noise quite a bit (at the cost of some sharpness) by turning off “Even Luma Peaking” and “Odd Luma Peaking”.