Interlaced video
Interlaced video
How are interlaced frames handled in NV? Do I have to separate fields before filtering and weave them back after filtering?
I have made tests filtering (very noisy) interlaced videos from old VHS tapes. It's possible just to ignore the interlacing and let NV filter the video. According to my tests, the result were not bad. Anyway, the filtered video looks much better than the original. But if the individual oven and odd filelds are studied, you can see the filtering artefacts.
If higher quality is required, the following workaround can be useful (until NV supports interlaced video). The idea here is to first separate the interlaced fields to form a new frame, perform filtering with NV and then combine the fields back to their original positions. The approach prevents the interference between the fields during the filtering, and it also allows the temporal filter to work without side effects. This is for VirtualDub version only.
Here it goes:
If higher quality is required, the following workaround can be useful (until NV supports interlaced video). The idea here is to first separate the interlaced fields to form a new frame, perform filtering with NV and then combine the fields back to their original positions. The approach prevents the interference between the fields during the filtering, and it also allows the temporal filter to work without side effects. This is for VirtualDub version only.
Here it goes:
- 1) In VirtualDub, add filter "deinterlace", and select mode "Unfold fields side-by-side".
2) Add filter "Neat Video"
3) Add "deinterlace" again, and select "Fold side-by-side fields together"
Code: Select all
720x576 1440x288deinterlace (mode: unfold)
1440x288 1440x288Neat Video
1440x288 720x576 deinterlace (mode:fold)
-
- Posts: 2
- Joined: Tue Jul 11, 2006 1:23 pm
- Location: Northern Lower Michigan, USA
Do you think field bobbing should be taken into account. In the past, when I've turned interlaced video into progressive, I've used VirtualDub's field bob filter with quarterline scan up/down or vice versa settings. This usually takes out any 'bobbing' up and down that the video does and make the frames "lin up". It seems that Neat Video's between frame filtering would benefit from the frames "lining up" by being able to keep more detail. However I'm not sure if that thinking is correct (maybe I need dome more coffee) I'll try to do some A/B comparison later. Can anyone comment on this?
This is a question of deinterlacing methods in general. I did not try the field bob filter for deinterlacing. The standard "deinterlace" filter in VirtualDub seems to work very well as it is and it provides good basis for accurate noise reduction in NV. If you try the field bob filter please let us know the results.
Thank you,
Vlad
Thank you,
Vlad
-
- Posts: 2
- Joined: Tue Jul 11, 2006 1:23 pm
- Location: Northern Lower Michigan, USA
Thanks for the reply. Well, I did a quick A/B experiment with the sample clip from the test kit.
For the first
Deinterlace - vdubs
NeatVideo
Weave them back together
For the second
Deinterlace - vdubs
Bob down/up
NeatVideo
Bob up/down (sometimes)
Weave them back together
On the second method, I tried different settings with the bob filter and with/without it again after NeatVideo, but no matter how I tried it either came out extra jumpy (when the bob settings were wrong) or it was slightly more blurry than method number one. I'd say Neatvideo does a great job with out the bob!
For the first
Deinterlace - vdubs
NeatVideo
Weave them back together
For the second
Deinterlace - vdubs
Bob down/up
NeatVideo
Bob up/down (sometimes)
Weave them back together
On the second method, I tried different settings with the bob filter and with/without it again after NeatVideo, but no matter how I tried it either came out extra jumpy (when the bob settings were wrong) or it was slightly more blurry than method number one. I'd say Neatvideo does a great job with out the bob!
That's rather disappointing. Blending together fields is not an alternative when working with video-based content.Working with interlaced frames is currently not supported in NV. I recommend to use progressive frames.
Timo's workaround (720x576 -> 1440x288 -> NV -> 720x576) is a nice idea, but the home-license's 720x576 pixel limitation doesn't allow me to use it.
Any chance of adding 1440x288 - compatibility to the home-license version ?
Thanks
Tobias
Actually, the current version (1.5) will work with interlaced footage but deinterlacing implicitly done by VD in this case is not so high quality as that provided by the dedicated "deinterlace" filter, so you may still want to explicitly deinterlace video sequence using "deinterlace".
Regarding the size limitations, please try to process the 1440x288 frames, it may still work in the Home plug-in.
Thank you,
Vlad
Regarding the size limitations, please try to process the 1440x288 frames, it may still work in the Home plug-in.
Thank you,
Vlad
right, but still not the best way to process e.g. sport recordings. For example I did a Tour de France video lately and be deinterlacing you're blurring the single frames extremely on the one hand while making the movement much less smooth (50fiels -> 25frames) on the other hand...Actually, the current version (1.5) will work with interlaced footage but deinterlacing implicitly done by VD in this case is not so high quality as that provided by the dedicated "deinterlace" filter, so you may still want to explicitly deinterlace video sequence using "deinterlace".
It gives me a error warning (limitation to 720x576), but after that it seems to the processing. Haven't rendered it to a new file yet though...Regarding the size limitations, please try to process the 1440x288 frames, it may still work in the Home plug-in.
Thanks,
Tobias
As far as I remember VD cannot provide interlaced frames to plug-ins in general. I believe interlaced video data is deinterlaced by VD before sending a frame to a plug-in so you have a choice "deinterlace using method A vs deinterlace using method B" rather than "deinterlace vs don't deinterlace". There is the method A - the implicit deinterlacing by VD itself, and method B - the explicit "deinterlace" filter. B provides better quality than A so it makes sense to use it.
Regarding the size warning, please try to render the clip anyway, it may still work with that size.
Thank you,
Vlad
Regarding the size warning, please try to render the clip anyway, it may still work with that size.
Thank you,
Vlad
Thanks for the input !
This week I'm working on VHS rips, so I'm going 360x576 now before splitting the fields, so it's ok anyway.
I finished the first two projects last night and the results are breathtaking ! I've put so much time into various AVISynth denoising filters and I really wonder why. While Neat is slow (3fps on my 3.4GHz machine) the results are the best I've ever seen in terms of denoising !
One more question: Currently I'm preparing a movie in which I have a hard time to find a 80x80 area without details.
Can I "get" the Neat settings before I resize the video (easier to find a 80x80 area in the 720x576 original) and then - afterwards - change the order of the VDub filters ?
In other words: is the example frame I pick used in the processing or is it just used to get the profile settings and I can disregard it once the profile is saved ???
THANKS !
Tobias
This week I'm working on VHS rips, so I'm going 360x576 now before splitting the fields, so it's ok anyway.
I finished the first two projects last night and the results are breathtaking ! I've put so much time into various AVISynth denoising filters and I really wonder why. While Neat is slow (3fps on my 3.4GHz machine) the results are the best I've ever seen in terms of denoising !
One more question: Currently I'm preparing a movie in which I have a hard time to find a 80x80 area without details.
Can I "get" the Neat settings before I resize the video (easier to find a 80x80 area in the 720x576 original) and then - afterwards - change the order of the VDub filters ?
In other words: is the example frame I pick used in the processing or is it just used to get the profile settings and I can disregard it once the profile is saved ???
THANKS !
Tobias
I am glad you are getting good results with NV.
Regarding finding a large enough area for analysis, resizing the video frame changes the noise properties so you have to be careful. For example, if you build a profile using a resized (larger) frame then the profile will describe the noise properties of that larger version only. The profile will make NV think that the noise elements are large (since they are resized too). And if you then use this profile to filter the original non-resized frames then filtration will not be very accurate because the noise elements in the original frames are smaller than in the frame you used for profiling. So, it is really important to build and apply noise profile to the same kind of frames: same size, same compression, same phase of post-processing workflow.
As far as I remember, you only need about a 64x64 area to build a profile. Probably it is a bit easier to find than 80x80. Also, if it is really difficult to find such an area in one clip, try to build a profile using a frame from another clip produced in the same or similar conditions. Do not forget to enable the "Adaptive filtration" option then.
Hope this helps.
Vlad
Regarding finding a large enough area for analysis, resizing the video frame changes the noise properties so you have to be careful. For example, if you build a profile using a resized (larger) frame then the profile will describe the noise properties of that larger version only. The profile will make NV think that the noise elements are large (since they are resized too). And if you then use this profile to filter the original non-resized frames then filtration will not be very accurate because the noise elements in the original frames are smaller than in the frame you used for profiling. So, it is really important to build and apply noise profile to the same kind of frames: same size, same compression, same phase of post-processing workflow.
As far as I remember, you only need about a 64x64 area to build a profile. Probably it is a bit easier to find than 80x80. Also, if it is really difficult to find such an area in one clip, try to build a profile using a frame from another clip produced in the same or similar conditions. Do not forget to enable the "Adaptive filtration" option then.
Hope this helps.
Vlad