DV capture quality sucks

[cooldude859]

Hi

I have a Panasonic GS85 and have been capturing via firewire. Aside from the expected interlace lines, I am getting some kind of vertical pattern that odesn't move with the video. It basically causes the video to look like hell. Even after using a high quality deinterlace in virtualdub, the verticals are still there. I have seemingly tried everything from different capture programs. Removing every possible DV codec and installing others. I even transferred a captured file to another pc and still see the problem.

But, my old pc has a capture card(composite) so I tried capturing that way. Obviously, not as sharp a picture but NO ARTIFACTS. Much more useable than the DV capture.

I don't really get it. I have seen others across the net complaing about the mysterious vertical lines but have found no solution to this. Is it caused by the camcorder?

I hope someone can solve this one. I like the crisp image DV gives. But I hate the artifacts.

Thanks for any help.

[edDV]

Hi

I have a Panasonic GS85 and have been capturing via firewire. Aside from the expected interlace lines, I am getting some kind of vertical pattern that odesn't move with the video. It basically causes the video to look like hell. Even after using a high quality deinterlace in virtualdub, the verticals are still there. I have seemingly tried everything from different capture programs. Removing every possible DV codec and installing others. I even transferred a captured file to another pc and still see the problem.

But, my old pc has a capture card(composite) so I tried capturing that way. Obviously, not as sharp a picture but NO ARTIFACTS. Much more useable than the DV capture.

I don't really get it. I have seen others across the net complaing about the mysterious vertical lines but have found no solution to this. Is it caused by the camcorder?

I hope someone can solve this one. I like the crisp image DV gives. But I hate the artifacts.

Thanks for any help.

Can you post a frame or two showing the problem?

Since the problem follows to a different computer, the camcorder seems to be the source of the problem. DV transfer should result in identical pixel data on the tape and hard drive file. Do you see any indication of the problem playing the tape from the camcorder to a TV monitor over S-Video?

[Baldrick]

Moving you to our dv forum section.

[cooldude859]

Here's a screen shot. The lines are more apparent on my widescreen LCD monitor. Not so apparent on my 15" CRT. But still there. The photo doesn't look as bad as the movement in the video.

Thanks.

[JohnnyMalaria]

Your image is 1280 x 960 and so it has been enlarged (presumably from playing full screen). The artifacts may or may not be due to that.

You need to capture a frame at its native resolution (720 x 480) and post that so we can see if the same artifacts are present. You should do that from VDub and from Windows Media Player to compare (they should use different decoders).

[filmboss80]

I think I know what the OP is talking about with the vertical banding, as I have seen it in dark scenes when working with Japanese footage (0 IRE?) in the DV format. I used both Panasonic and Cedocida codecs and got the same results. The problem worsened with title keys and effects (recompression) in the editor's timeline. Never figured it out. I was thinking maybe the OP got a camcorder with the black level set low. Can't be sure if that would affect anything. Maybe some of you can enlighten us.

[2Bdecided]

Is it around colours and colour boundaries, looking like a kind of mesh when there is movement?

If so, it's probably NTSC DV 4:1:1 chroma, and the (bad) way it's being decoded on your machine.

The composite capture just smears it.

Just a suggestion - it might not be that.

Cheers,
David.

[edDV]

VLC is also good at snapshot directly from the DV-AVI file. It would be a good crosscheck on any artifacts added by Vdub DV codecs. Snapshot is under the "Video" tab.

Also VLC is good for deinterlacing DV playback. You will also find "Deinterlace" under the "Video" tab. "Disable" shows pure 480i (interlace). Try "mean" for 720x480p deinterlaced display. Below you see the results on the horizontal stock ticker.

480i direct DV-AVI


VLC "mean" deinterlace from DV-AVI playback.

[cooldude859]

Hey thanks for all that helpful info. After some careful reading I stumbled onto trying to "properly" deinterlace the video. At this point I was willing to try everything. I found a really neat tutorial for using Twixtor and FieldsKit to fix the video. I was skeptical but it worked very nicely. Surprisingly, it removed all the artifacts and interlacing without destroying any of the quality. It also converted the 60i>60p>24p with amazing results.

Sorry if I sound like a salesman but I feel a huge sigh of relief considering the hours of frustration. I am on my old pc so I can't post the capture, yet, but will try over the weekend. It really does look a lot more like film to me.

[filmboss80]

Hmm... Your initial description was of a vertical pattern. Deinterlacing plug-ins like FieldsKit usually remedy horizontal interlace artifacts. I, too, would be real interested in seeing a good quality screen shot of your original problem.

[JohnnyMalaria]

VLC is also good at snapshot directly from the DV-AVI file. It would be a good crosscheck on any artifacts added by Vdub DV codecs. Snapshot is under the "Video" tab.

Also VLC is good for deinterlacing DV playback. You will also find "Deinterlace" under the "Video" tab. "Disable" shows pure 480i (interlace). Try "mean" for 720x480p deinterlaced display. Below you see the results on the horizontal stock ticker.

480i direct DV-AVI


VLC "mean" deinterlace from DV-AVI playback.

Hmmm. Notice how the deinterlacing has introduced vertical banding - e.g., the edges of the yellow diamond in the Brazilian flag.

[aedipuss]

doesn't look like banding to me. more like typical square pixel diagonal line jaggies. it's compounded with deinterlacing as you've only got every other line, leaving out the smoothing of the in between lines. you can see it in the yellow stars also.

[JohnnyMalaria]

I don't agree

Look at these two samples. After deinterlacing, there are notable vertical bands (e.g., the upper horizontal edges of the star) that are more than two lines tall even in a region of the image where there are no interlacing artifacts.

[aedipuss]

your eyes must be better than mine, all i see are jaggies above and below the diagonal lines and no banding in the center of the star where there are no lines. the jaggies may join together where they are close, and look like banding. but i think it's just where the red pixels intruded into the yellow pixels at the corners of the squares. and the same with the yellow into the red areas. but what do i know

[JohnnyMalaria]

I guess using MS Paint in a hurry didn't help.

Here's a different take:



As it happens, a couple of days ago I did notice this when play a DV file in WMP. I can't recall the details though (which PC, OS version, display etc etc).

[jagabo]

I guess using MS Paint in a hurry didn't help.

Here's a different take:...

That's the mark of a DV decoder which is duplicating the U and V samples when expanding from 4:1:1 to 4:4:4. A filter like 4:1:1 Helper can fix that. Or using a DV decoder that does a smooth interpolation.

Here a 4x enlargement (nearest neighbor) of a crop from a DV AVI before and after 4:1:1 Helper (in VirtualDub):

[JohnnyMalaria]

But in edDV's example, aren't both images decoded with the same DV decoder but in one case displayed as-is and in the other the already decoded image is deinterlaced by VLC?

If it were just 4:1:1 sampling, I'd expect to see it both images and the bands should be 4 pixels wide, not 2.

[edDV]

But in edDV's example, aren't both images decoded with the same DV decoder but in one case displayed as-is and in the other the already decoded image is deinterlaced by VLC?

If it were just 4:1:1 sampling, I'd expect to see it both images and the bands should be 4 pixels wide, not 2.

It was intended as a quick illustration of playback deinterlace. VLC playback deinterlace methods are described here
http://wiki.videolan.org/Deinterlacing

The source was 524x480i 4:2:0 MPeg2 cable (comcast cable box) captured S-Video to a Canopus ADVC-100 (analog to DV conversion) then played with VLC from the DV-AVI file.

A better test of the VLC deinterlace would be from a camcorder capture.

[vhelp]

These are just theories mind you, because the true origin is not known, and the steps that
lead to the final posted image are also not fully known.

The images that edDV posted are a bit hard to decirn because they seem to be from an analog
type source and they are pretty blurry. That indicates to me some equipment filtering had been an
internal part of the system or user-applied. Never the less, you can't really completely rule one thing
here because anything that comes from broadcast (tv) are almost always full of errors of one sort of
another. So, everything else following that fact are pointless.

The other factor is relying on a softwares filter function. Personally, WMP and VLC do not quite appeal
to me, and I wouldn't rely on certain features of them when analizing a video or frame is important.

As an example, that pic he posted could already have sub-sampling errors from within the source level
before he captured it. Plus, noisy interlaced sources are usually desceaving in filter outputs.. ie,
the deinterlace function. Besides, he was just posting a deinterlace example

Now, given the facts of todays software and their upgrade/version issues, if I needed to export a
single frame from a video I would perform this function from my own image tools and just skip these
known applications (WMP and VLC are examples) all to gether and review/analize them for my own
purposes.

edit: I didn't read the whole discussion, so I missed jagabo's and edDV's followup
posts. Sorry bout that fellows

-vhelp 4687

[jagabo]

But in edDV's example, aren't both images decoded with the same DV decoder but in one case displayed as-is and in the other the already decoded image is deinterlaced by VLC?

If it were just 4:1:1 sampling, I'd expect to see it both images and the bands should be 4 pixels wide, not 2.

Yes, you are right. After a closer examination, the light and dark streaks are 2 pixels wide and the pattern repeats every 4 pixels. If you take the original picture and split it into two fields (unfold fields side-by-side in VirtualDub, or SeparateFields() in AviSynth) you'll see that the separated fields have the light/dark streaks -- the weaving of the two fields in the original PNG masks this effect. I suspect they are related to NTSC dot crawl.

I think vhelp is right, the images we see are a single frame far removed from the real source. It's hard to say anything definitive about the causes of the artifacts. Hey, edDV, if you still have the DV AVI could you upload a second or so of it?

By the way, the left part of the image I posted earlier was decoded from DV to RGB by ffdshow (grabbed with VirtualDub's Video -> Copy source frame to clipboard). Using Cedocida instead gave a smooth result like the right part of the image.

[Undead Sega]

what is 4:1:1 Helper???

[jagabo]

4:1:1 Helper is a filter for VirtualDub:

http://www.geocities.com/xesdeeni2001/

[Undead Sega]

cool, but what does it to exactly? looks interesting.

[jagabo]

NTSC DV subsamples color at 1/4 the resolution (on the horizontal dimension) of the brightness resolution. In other words, the luma channel is 720x480 but the chroma channels are only 180x480.

When converting this to RGB some programs resize the 180x480 chroma planes using a simple nearest neighbor (aka point) resizing. 411 Helper fixes that by smoothly interpolating the colors channels.

[Undead Sega]

ohh right, does this also go for PAL DV?

[JohnnyMalaria]

PAL DV is sampled as 4:2:0 which means for every two horizontal luma pixels there are only one of each chroma pixel and likewise for every two vertical luma pixels.

In other words, for a 2 x 2 block there are 4 luma pixels but only of each chroma.

4:2:0 sampling is also used for PAL and NTSC DVD.

[Undead Sega]

so really, there isnt a need for 4:1:1 Helper for PAL DV right?

also, if u had an NTSC DVD that is sampled at 4:1:1 (dont know if u could tell), would this filter come in handy?

[JohnnyMalaria]

so really, there isnt a need for 4:1:1 Helper for PAL DV right?

also, if u had an NTSC DVD that is sampled at 4:1:1 (dont know if u could tell), would this filter come in handy?

There may be a 4:2:0 Helper for PAL - I don't know.

NTSC DVD is always 4:2:0 and never 4:1:1.

[Undead Sega]

There may be a 4:2:0 Helper for PAL - I don't know.

cant u still use 4:1:1 Helper on PAL DV footage? would it work or give some kind of result needed?

[edDV]

PAL DVCPro uses 4:1:1 sampling. 4:1:1 has less generational loss than 4:2:0 which requires a spatial chroma interpolation each time it is recoded. 4:1:1 better fits pro workflows that can can go 3 or more recodes to finished product. HDCAM use 3:1:1 sampling for similar reasons.

4:1:1 disadvantages show when sharp chroma transitions happen combined with relatively flat luminance as shown in jagabo's pattern above. Under similar circumstances 4:2:0 breaks into 2x2 blocks.

4:1:1 helper wouldn't match the 4:2:0 issues.