|
|
| miamicanes |
| Posted: Feb 22 2003, 07:02 PM |
 |
|
Unregistered
|
Is it better to apply a good deinterlacing algorithm first, then apply noise-reduction algorithms to the pseudo-progressive source that results? Or is it better to do the noise-reduction first, then deinterlace?
Conceptually, it seems like better results would be possible from an arbitrary noise-reduction filter if it were applied to the pseudo-progressive source generated by a superior deinterlacing algorithm, because most noise reduction filters otherwise seem to end up implementing their own (probably inferior, at best equally effective) deinterlacing algorithms anyway.
In other words, it seems like the most efficient, and probably most effective noise-reduction strategy for source captured from VHS to HuffyUV at 29.97 frames/second interlaced, would be:
* convert it to 59.94 frame/second progressive using SmartDeinterlace with settings optimized for the specific video content
* apply noise-reduction filters (Dynamic Noise Reduction, SmartSmoother HiQ, 2d Cleaner, etc) to the pseudo-progressive source
* throw away alternating odd and even fields of the final 59.94 frame/second progressive video to produce 59.94 field/second interlaced video.
Is this a valid assumption? Is it likely that something like SmartDeinterlace could produce better source for the noise-reduction filters to tackle than they'd likely generate on their own internally?
I experimented a little yesterday, but screwed up and ended up turning the 29.97frame/second source captured from VCR into 29.97 frame/second progressive by mistake (which just made things worse and caused judder). Actually, that raises another issue...
HOW DO you use SmartDeinterlace to turn interlaced video of a particular FIELD rate into progressive video of the same FRAME rate (vs keeping the frame rate constant and just removing the artifacts)? I hope I'm wrong, but it almost looks like all the deinterlacing tools are specifically designed to keep the framerate constant (for MPEG-1 VCD purposes) rather than turn 59.94fps into 59.94 field/sec. |
|
| fccHandler |
| Posted: Feb 22 2003, 07:29 PM |
|
|
Administrator n00b
Group: Moderators
Posts: 3961
Member No.: 280
Joined: 13-September 02
|
Hello.
You asked some good questions, but I'm a little confused about what kind of video you are trying to produce. Do you intend your final video to be interlaced? Or are you wanting to convert true interlaced source to progressive?
| QUOTE | | * convert it to 59.94 frame/second progressive using SmartDeinterlace with settings optimized for the specific video content |
SmartDeinterlace doesn't do a conversion to 59.94 fps progressive. All it does is blend (or interpolate) fields in each frame of the video.
| QUOTE | | * throw away alternating odd and even fields of the final 59.94 frame/second progressive video to produce 59.94 field/second interlaced video. |
Huh? You lost me there.
Take a look at this Web site, it's loaded with info about deinterlacing:
http://www.100fps.com
--------------------
May the FOURCC be with you... |
|
| miamicanes |
| Posted: Feb 22 2003, 07:58 PM |
|
|
Unregistered
|
I'm ultimately capturing interlaced video for authoring interlaced DVDs.
The main rationale behind deinterlacing before filtering is that progressive video is infinitely easier to manipulate than interlaced video. To make intelligent decisions, a filter ideally needs to know what a scanline's neighbors (from the previous and following field) are doing too. In other words, a filter operating on scanline #100 (among a field of other even scanlines) ideally needs to know what happened to scanlines 99 and 101 in the previous and following fields of odd scanlines. If it ignores activity in scanlines not part of the current fieldset (looking only at scanlines 98 and 102, for instance), it risks ignoring important information that might help it make better decisions with regard to the current scanline of interest.
As I see it, there are basically two ways this can happen: you can apply an optimized deinterlacing algorithm to the entire 59.94 field/second 29.97 frame/second interlaced source first, and run the filters on the 59.94 field/second 59.94 frame/second progressive video that results... or you can leave it up to the filter to do the same thing over and over and over and over again, probably with inferior results. If you're applying only a single filter witha single pass, it's probably more efficient to just let the filter handle the deinterlacing on an ad-hoc basis. On the other hand, if you're applying a CHAIN of filters, or one or more filters needs to make multiple passes, deinterlacing once and deinterlacing right should theoretically produce better (or at least faster) results.
Ultimately, of course, the nice, clean, filtered 59.94 frame/second progressive source needs to get butchered back into 59.94 field/second 29.97 frame/second interlaced video to burn to the DVD. But the key point is that deinterlacing it first and applying filters to the intermediate progressive source (even if we're ultimately going to throw away half the scanlines when rendering the final video to DVD) should still save time and produce better results, with the only real negative consequence being that the amount of disk space required to hold the intermediate workfiles would probably double or quadruple.
Think of the preliminary deinterlacing to progressive as kind of a cacheing scheme to save filters down the chain from having to do it all over again.
I think the thing that might be confusing is that lots of people try to deinterlace 59.94 field/29.97 frame interlaced video to 29.97 field/29.97 frame progressive (which is precisely what MPEG-1 wants). Doing THAT will cause judder if it's ultimately turned back into 59.94 field/29.97 frame video at some later point, because half the TEMPORAL info will have been discarded/mangled along the way. HOWEVER, by deinterlacing 59.94 field/29.97 frame to 59.94 field/frame, you're preserving all the temporal information as well, even if DVD can't handle that particular flavor of MPEG-2 natively (at least, not yet).
Getting back to the second part of my post, the main thing I need to determine is whether VirtualDub/Smart Deinterlace CAN'T deinterlace 59.94 field/29.97 frame to 59.94 field/frame progressive, or whether it's just something that nobody has done and written about doing yet because it would produce progressive video that would currently be useful only as an intermediate format (or for viewing on computers). |
|
| fccHandler |
| Posted: Feb 22 2003, 08:16 PM |
|
|
Administrator n00b
Group: Moderators
Posts: 3961
Member No.: 280
Joined: 13-September 02
|
You shouldn't ever need to deinterlace if you're making DVDs.
Here is what I would try in the filter chain:
- Filter: Deinterlace (Unfold fields side-by-side)
- (noise reduction filters go here)
- Filter: Deinterlace (Fold side-by-side fields together)
That way, NR filters get to work on the proper fields, though there may be some bleeding along the center edge with 2D NR. (I guess you could crop it away.)
Hope this helps.
--------------------
May the FOURCC be with you... |
|
| ADVC100 |
| Posted: Feb 27 2003, 03:29 AM |
|
|
Unregistered
|
The problem I've found with using the unfold-filter-refold method is that the filter is working on an incorrect aspect ratio, meaning any 2d blurring that happens during the noise filtering is vertically stretched when the two fields are refolded together.
More and more filters are including an "interlaced source" option into the filter, one of my favourite noise reduction filters is the Smart Smoother by Donald Graft, just wish the VHS filter had an "interlaced source" option too. |
|
| David.Bucci |
| Posted: Feb 27 2003, 06:15 AM |
|
|
Unregistered
|
Just an fyi, Don Graft's "Smart Bob" will produce the 59.94 fps progressive video you're looking to make in the beginning of your discussion. |
|
|
