|
|
| davexnet |
| Posted: Feb 28 2011, 07:49 PM |
 |
|
Member
Group: Members
Posts: 16
Member No.: 25734
Joined: 25-June 09
|
Hi all,
I'm familiar with spacial and temporal noise reduction as used in most noise reduction filter.
However, the MSU 2.51 noise reduction filter has a Bilateral mode which uses spacial and photometric.
I looked for an explanation on the web in plain English as to how this NR works,
but all I found is a load of mathematical formulas and I was looking for an overview
of the theory behind it.
Anybody heard of it or tried this filter? I've had some interesting results with it.
Quite close to NeatVideo (demo) but much faster. |
|
| phaeron |
| Posted: Mar 5 2011, 10:10 PM |
|
|
Virtualdub Developer
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02
|
Bilateral filters are the algorithm that most spatial NR filters use -- it's just a weighted average of a neighborhood of samples within the same frame, where the weights are determined by color distances. I don't know what photometric mode does, but it may involve a change in the weight computation. In any case, the mathematical formulas are probably the description you're looking for. Many filters of this type can be described in a single equation.
|
|
| davexnet |
| Posted: Mar 7 2011, 01:32 AM |
|
|
Member
Group: Members
Posts: 16
Member No.: 25734
Joined: 25-June 09
|
Thanks Phaeron, I was just wondering if this mode of NR employed something similar to temporal NR,
where consecutive frames are considered in reducing the noise.
If not, it's quite impressive for a 2d filter. |
|
| evropej |
| Posted: Mar 7 2011, 05:18 PM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
Try this out for noise reduction
Add three smoother filters instead of the MSU denoiser
smoother 20
smoother 20
smoother 20
I found this method more successful than MSU for really grainy videos.
There are other temporal denoisers as well but the detail reduction is to severe.
You can optimize the filter as well by making sure your video dimensions are above 800 pixels.
Add a resize filter with lanczos resize and MSU deblocking for best results.
filter list:
MSU deblocking
resize ( 1000x1000 pixels for example ) lanczos
MSU deblocking
smoother 20
smoother 20
smoother 20
Last but not least, usually some histogram tinkering or gamma will reduce a lot of noise.
Cameras will bump up gamma to get that detail so by lowering the gamma, you reduce the noise. |
|
| JAFO |
 Posted: Dec 1 2013, 12:27 AM |
|
|
Newbie
Group: Members
Posts: 3
Member No.: 37515
Joined: 30-November 13
|
Hey everybody, my first post here.
I have been using VD for years but am far from being a professional. I do however try everything (including the kitchen sink if possible) to find remedies to problems.
The topic is noise reduction here so I'll inject something that may at first seem 'odd'.
Anyway try this stack on noisy video...
Field Bob (smooth/smooth)
Rotate (left 90)
Field Bob (smooth/smooth)
Deinterlace (mode: blend,keep bottom)
Rotate (Right 90)
Field Bob (smooth/smooth)
If its not enough- repeat...
I have used this stack on some pretty worthless/seemingly hopeless video's. It works amazingly on both noise and blocking artifacts, the downside is that some very slight blurring occurs but I have never had any pronounced motion blurs(even at 20reps). Saturation may also be enhanced but usually easily correctable. Chroma/Dot crawl (if present) may need to be addressed in advance as this may get enhanced in the process.
Try it I think you will like it.  |
|
| evropej |
| Posted: Dec 6 2013, 07:10 PM |
|
|
Advanced Member
Group: Members
Posts: 514
Member No.: 26523
Joined: 28-November 09
|
So since this topic has come up again, I will say something else.
Watching movies on the computer is not indicative of the quality you will see on a tv.
The typical television has a finite response time which acts as a natural filter for noise.
However, with newer tvs, the response frequency is being raised.
I find this truly amazing given the fact that most movies are recorded at much lower frequencies.
So what they are doing with higher response times is simply allowing more noise to show up.
Not only that but the interpolation is even worse which creates more problems.
For the best noise removal plugin, nothing really beats neatvideo which is not free by the way.
I have tried many and many methods or combinations of filters as suggested above.
Inherently, detail resides in high frequency and so does noise.
So removing noise will result in some removal of detail which causes blurring. |
|
