Capture Quality 1394 vs MJPEG

[Ted222]

I was wondering if I could get some opinions about the quality of video captured to a computer to edit. Which is better?
I have tried 1394 and S-Video using a mini-DV camcorder and a Hi-8 Camcorder using PICVIDEO, MJPEG, and default DV.
My results are:
PICVIDEO - high quality but editing not great
MJPEG - high quality but file size is huge
DV(1394) - good quality but not as good as the others and jittery video sometimes.

G450eTV
P3-1000
45 gig
60 gig video raid (promise 100)
384 RAM
Intel 815e MOBO
Canon Optura PI
Sony TRV99 Hi8
Media Studio Pro & Video Studio 4 SE

Thanks,
Ted


[This message has been edited by Ted222 (edited 10 June 2001).]

[Dr Mordrid]

I have 3 mixed mode editing systems here.

I find DV fine for generic footage, but it's inferior for encoding to MPEG or when doing multilayered special f/x composites or bluescreening. For these I use HuffYUV captures.

This is a complex topic, but the issue is mainly because of DV's reduced colorspace vs. analog video. As such if I am doing a planned project I use analog for the multi-layered f/x and bluescreen scenes.

Any such material gets exported into HuffYUV before their inclusion in projects of another format. This way the full project export is the first time the effects get compressed by any lossy compressor (basically anything but HuffYUV).

This reduces artifacting considerably.

Dr. Mordrid



[This message has been edited by Dr Mordrid (edited 11 June 2001).]

[Ted222]

Dr. Mordrid,
I will try HuffyYUV and post the results if this thread is still active.

Ted

[wkulecz]

If your "DV is jittery" is from the DVD max dualhead feature and MSP6 I find MSP6 just doesn't work right with DVDmax dualhead output for DV. Premiere6 on the same system works great with DVDmax dualhead even when also sending DV to the camcorder at the same time.

I've verified this on W2k with two very different systems Athelon-700 and PIII-800.

--wally.

[Railie]

Capture from DVcam, using S-video? I mean that DV is DV - you get in PC what you have in your tape. Is my technique bad, when I capture to DV(it is easy-start/stop/segments)and then convert segments for FX to HUFF?
My final work is tored as DV - eport to tape, and also for creating SVCD. Is good idea, for creating SVCD, export my final video to HUFF before encoding to MPEG2?
Now I encode directly from DV, using interpolate YUV data from 4:1:1 to 4:4:4.

Thanks Railie

[Frank Marshall]

Hey Doc,

About analogue footage encoding to MPEG better than DV:

"I find DV fine for generic footage, but it's inferior for encoding to MPEG..."

Are you thinking here of high quality analogue sources such as Betacam SP or U-matic? What I'd like to know is whether this assertion holds true when comparing Hi8 sources with DV for MPEG encoding.

Thanks.

[This message has been edited by Frank Marshall (edited 18 June 2001).]

[Dr Mordrid]

I've capture Hi8 both as composite and S-Video and either beats DV when it comes to artifacts in the final MPEG hands down.

In actuality the issue comes down to;

1. DCT compressed or not

2. colorspace

Uncompressed sources, or those that have only been Huffman encoded like HuffYUV, have no quantization vs. their original 4:2:2 source material. This provides a wider dynamic range and prevents a lot of recompression related artifacts.

With DCT compressed video (MJPeg, DV etc.) this is not the case.

Also, DV's reduced 4:1:1 colorspace has a large negative effect when it comes to quantization related artifact generation in solid color and shadow areas. Much of the problem is actually due to a double quantization when encoding DV to MPEG. First it gets quantized to 4:1:1 in the camera, then to MPEG's 4:2:0 when the MPEG encoder does its work.

In these operations a lot of pixel averaging gets done, and this can cause many of MPEG's 8x8 macroblocks to end up with a solid color throughout. Not good.

Generate enough of these solid color macroblocks and you end up with the very visible block artifacts so many people experience with DV footage. Again, its a matter of a reduced dynamic range in the color content.

Of course the generation of large numbers of same-or-similar colored macroblocks is more likely in single color, shadowed or poorly lit regions.

Because of these problems the best DV footage to use for MPEG is that which is;

1. well and evenly lit

2. varigated. Meaning few large, solid color areas.

3. to be rendered in full resolution or close to it. scaling down to low resolutions (ex: VCD) makes matters worse.

As a result I usually have a Hi8 with me just in case I run into a situation that would be problematic. When doing studio footage destined for keying the captures are always done with the Hi8 feeding the capture board directly by S-Video for highest quality.

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 18 June 2001).]

[mikiem]

Hi

Doc's absolutely right.

Just another way to think of it, in consumer type DV, the amount of picture data controls everything else. Record something with a lot of motion and colors, and it will automatically be compressed more then a talking head. And of course, everything will be compressed by the camera's circuitry. Any further conversion will hurt some.

An analog tape is more or less just like a tape recording of your voice -- just puts down what it hears with quality dependent on equipment. You control any compression when you capture/edit the video.

For what it's worth, while there is a huge amount of stuff written about DV, might be surprised to know that FlatEarth has supplemented shots with Hi-8, & Lucas at one time at least I think was using it for instant playback/previews on the set.

mike

[Dr Mordrid]

I believe I heard something similar.

These problems will remain until 4:2:2 DV becomes widespread instead of being limited to propriatory formats.

IMHO the biggest mistakes made in setting the DV standard were;

1. using 4:1:1 to begin with

2. doing the 4:1:1 quantization BEFORE the Huffman encoding instead of after.

Duhhh....

Dr. Mordrid

[Dr Mordrid]

The actual key to getting 4:2:2 DV is the bitrate. 4:1:1 DV is 25 mbps while 4:2:2 DV is 50 mbps.

For future reference;

The video systems using 4:2:2 are D-1, D-5, DigiBeta, BetaSX, Digital-S (aka: D-9) and DVCPRO50

The video systems using 4:1:1 are NTSC DV, DVCAM & DVCPRO

The video systems using 4:2:0 are PAL DV, DVD & MPEG-2 MP@ML

Now...it would seem that PAL DV would be of advantage for generating DVD/MPEG-2 because of the matching colorspaces, right? Wrong.

PAL DV @ 4:2:0 has problems with multigeneration losses both in editing and rendering into other formats. This is specifically why Panasonic developed PAL DVCPRO using 4:1:1.

The problem is that while it's sampled at half the horizontal luma rate, just as 4:2:2 is, it's also sampled at half the VERTICAL rate. This means a color sample for every other scan line, not every scan line as with 4:1:1 or 4:2:2.

Uppance:

If your target is purely professional then use 50 mbps 4:2:2 if you can afford it. This will allow you to do higher quality MPEG's with much less in the way of complexity. Just be sure your editor can handle the format you choose.

Otherwise, learn to intersperce analog with the DV footage where necessary in a mixed mode system. This is far more economical and practical for most of us.

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 19 June 2001).]

[JerryH]

Quick question Doc, I've seen the 4:1:1 and 4:2:2 ratios kicked around in this thread; are these also sometimes referred to as DV type 1 and DV type 2, or is that something else altogether?

[Dr Mordrid]

Something else altogether. They are separate and distinct things.

DV type 1 & 2 refer to how the DV audio is encoded and if it's OpenDML compatable. Here's a link explaing them;

http://www.puremotion.com/videoediti...l/dvfiletypes/

Now, about colorspace ratios....

4:4:4, 4:2:2, 4:1:1, 4:2:0 etc. refer to the "colorspace" of the video data. Lets assume a 720x480 video frame. Also understand that there are three components to a typical analog video signal. These can be either YUV or RGB:

Y = luma = B&W portion of the signal
U = chroma R = half the color signal
V = chroma B = half the color signal

or

R = red
G = green
G = blue

Most often video uses a YUV signal, the complete color range of which can be obtained mathematically from the above parts. You can also convert YUV to RGB and back. We'll avoid the matrix math calculations for now

In each of the colorspace ratios each component is sampled at a different rate. In all cases YUV's Y (B&W) value is sampled at a full resolution of 720 horizontally.

4:4:4 = every value sampled horizontally and vertically. Think in terms of a full 720x480 rez for both B&W and color information.

4:2:2 = every other chroma value sampled horizontally. Think in terms of a full 720 for B&W, but only 360 samples for color. All scanlines sampled.

4:1:1 = every fouth chroma value sampled horizontally. Think in terms of a full 720 for B&W, but only 180 samples for color. All scanlines sampled.

This low horizontal chroma (color) sampling rate can (and often does) cause block artifacts at the edge of overlay objects and when doing bluescreens. 4:1:1 can be used for these effects, but it takes a lot of work and careful attention to detail. Even then....

4:2:0 = nonstandard definition. Every other chroma value is sampled horizontally. Think in terms of a full 720 samples for B&W and 360 samples for color like with 4:2:2.

The rub: only every other scanline of chroma information is sampled! This can cause problems when rendering effects as it reduces multigenerational quality.

Basically the more samples the better with 4:4:4 being the best found in most medium to low end gear.

Ex: the RT-2000 takes this a step further by using 4:4:4:4 RGBA internally, with the fourth "4" being the addition of an alpha channel for higher quality overlays.

Full shot studio gear can go to 8:8:8:8 using either YUV or RGB.

Hows that?

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 19 June 2001).]

[wkulecz]

NO, DV type 1 and DV type 2 have nothing to do with the colorspace (4:1:1 or 4:2:2).

DV Type 2 is the original Video For Windows AVI file format with a DV codec and possibly "enhanced" for DirectShow to overcome the 2/4 GB FAT32 limits.

DV Type 1 is a new AVI format, I believe introduced with DirectShow, its pretty much the DV stream as recieved/sent to the camcorder.

Basically Media Studio Pro 6 works best with DV type 1. Premiere 6 works best with DV type 2. Until I realized this I thought I'd wasted my money on the Premiere 6 update. This fact makes it rather inconvienent to do parts of a project in MSP6 and other parts in Premiere6. Eventually I'll probably move to Premiere6 unless Ulead comes up with a fix for the poor performance when DVDmax is enabled on my G450.

--wally.

[JerryH]

thanks for the info, folks!

And Doc, your info has been enlightening, but disheartening at the same time. I have been seeing the artifact stuff to which you refer (I have a D8 camera and Pyro firewire card) after rendering. I thought perhaps I had some bad tape or the camcorder's heads needed cleaning. Ah well, guess I'll just need to glom onto a RT or something....of course, the wife probably wouldn't dig that.

[Dr Mordrid]

Well...remember that the RT2x00's also use DV primarily. Its big advantage, other than realtime, is its compositing engine. It upsamples the the DV footage to 4:4:4:4 and adds an alpha channel for internal use on the fly.

This is used not only for compositing in realtime but for renedered effects that don't meet the realtime limitations (ex: 3 or less layers etc.). This really improves the result.

In most editing software you would use with OHCI cards DV is upsampled to 4:2:2 for compositing, which isn't up to the RT's 4:4:4:4 RGBA trick. Also, no alpha channel is applied. That alpha channel in the video file itself makes the composites and effects much cleaner.

Dr. Mordrid



[This message has been edited by Dr Mordrid (edited 19 June 2001).]