Optimal resolution for capturing video from TV?

[DAR]

Several weeks ago there was a very informative thread discussing the best capture resolution for capturing video from VHS tape (capture using Marrox Marvel MJPEG).

Is there an optimal reolution setting for capturing from TV? By optimal I mean the minimum resolution setting that captures all of the information available from the TV signal.

I'm relatively new at this, so sorry if this question is not relevant or somehow self-evident.

[Dr Mordrid]

352x480 gives you all the pertinent data required for "natural" video sequences.

Natural video is, in general terms, scenes without graphical titles, keyed and overlay effects or graphics. Basic talking heads, informal family events and elementary school functions are typical of this type.

For video where you expect to use graphical title sequences, keyed or overlay effects go with 704x480 because of the higher linear resolution. This helps a lot in getting high quality out of the graphical portions.

This catagory of events would include sports and formal events like weddings, religious cerimonies (bas/bar mitzvah's, confirmations etc) or fake newscasts. Organic wipes, keyed effects and graphics overlays are very useful for enhancing these events if you use them judiciously. And that is a BIG if.

If you decide to cross boundaries, such as by making a family Christmas video in the Spielberg/Cameron model, then 704x480 would apply there as well.

Dr. Mordrid



[This message has been edited by Dr Mordrid (edited 28 September 2000).]

[Biker]

Hey Doc I thought we needed 768x576 (PAL) to capture all the info, or are a lot off these 'extra' lines just for 'hidden' info IE. Teletext services and subtitle information?

Any advice appreciated.

------------------

[Dr Mordrid]

Full D1 for NTSC is 720x480. The 768 width frames include even more overscan than this frame does. Since you don't normally see the overscan you don't really gain much by using it. The Marvels use only 704x480 and do just fine as again any extra is in the overscan area and invisible on properly adjusted sets.

Teletext & Closed Captioning are actually contained in an area between the video frames, not within them, so they aren't reflected in the frame sizes.

Now for why "natural" video can be used at lower resolutions. This is a bit complex so I'm going to try and simplify the rest a bit to make it easier to understand. I'm only going to speak to NTSC as adding too much about PAL would just make things more complex without changing the fundamentals. Techies forgive me;

The raw signal captured by most NTSC analog cards is YCrCb, commonly (but incorrectly) known in video circles as NTSC "YUV".

YUV is actually a term used in PAL systems defined by the EBU (European Broadcast Union). Still, analog video is very commonly referred to as YUV in both systems.

In YCrCb the Y component refers to the luma (brightness) portion of the signal. The Cr and Cb components contain the chroma (color) portions.

The colorspace of a given NTSC analog video signal is expressed numerically using the following format: nY:nCr:nCb, where n is a value from 0 to 4. YCrCb video with full resolution sampling of both luma and chroma would be expressed as 4:4:4.

4:4:4 sampling is not often used at the consumer or prosumer level. Instead, a slightly reduced colorspace of 4:2:2 is most often used. The main reason is that 4:2:2 offers a bit of compression at little loss in visual quality. This compression is at the hub of why "natural video" can be captured at 352x480 without losing much quality.

In 4:2:2 each pixel of the luma is sampled but each chroma sample is spread over two pixels. Because of this the luma portion of the signal contains most of the linear detail of the scene. The chroma portion offers half the horizontal resolution of the luma portion.

Having half resolution for the color portion of an image is nothing unusual for humans as this is how our own visual systems work. Humans normally see with much higher resolution in B&W than with color because we have more B&W sensors (rods) than color sensors (cones) in our retinas. As such our visual systems "prefer" the luma portion of a video signal when it comes to viewing its details.

Putting this all together:

In a full width (704x480) frame you would be getting 704 units of luma detail but only 352 units of chroma. Great if you have highly detailed graphics, titles or animations in the frame. Largely unnecessary if you're shooting live action absent graphics.

In a half width (352x480) frame the luma resolultion would be halved, but you'd still be getting all 352 units of chroma. Absent any highly detailed graphics, titles or animations in the frame this is all you need.

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 02 October 2000).]

[Dr Mordrid]

When YCrCb is resampled to a 352 wide frame all that's necessary is to drop half of the horizontal pixels. Since each chroma value is duplicated over two horizontal pixels in the full size frame no unique chroma information is lost as long as you are consistant about which ones are clipped.

If you were to select a 176 frame width then you would be dropping chromas whose information is unique. In this case you would lose horizontal chroma resolution.

Dr. Mordrid


[This message has been edited by Dr Mordrid (edited 02 October 2000).]

[Michel Carleer]

I agree that it is not necessary to remove anything to the chroma data, but if you don't, the format of your file is no longer 4:2:2, as you have the same number of Cr, Cb and Y. Rather, it is 4:4:4. Which is a perfectly valid format, but I am not sure what the resampling programs (or the digitizing chips) do. You see, you can do either: undersample the chroma as well as Y and the resulting file is also 4:2:2, or retain all chroma and end up with a 4:4:4 file.
Michka

[Michel Carleer]

Dr. Mordrid,
Withoutoffense, one additional comment (on your otherwise excellent explanation on the YUV video format) and one question:

One additional reason why it is perfectly acceptable to digitize in YUV 4:2:2 is that a video signal (at least the composite one and, I suspect, S-video as well) transmits chroma with a far lower resolution than the luma (1 MHz versus 5 MHz). The lower chroma resolution has been chosen for the reason you give: the human eye is far less sensitive to chroma res.

Now for the question: are you sure that a picture sampled at half res (352x480) still contains 352 chroma values? I bet it is more like 176. After all, 4:2:2 is always 4:2:2, no matter what the horizontal res.
Michka


[This message has been edited by Michel Carleer (edited 02 October 2000).]

[Biker]

Thanks Doc and Michka, for this info/discussion.

Out of interest does this still apply for PAL systems IE. The chroma sampled over 2 pixels? I am aware of the Basic difference in their operation as a system.

Thanks Guys - Biker

------------------

[chm]

For additional clarification on the meanings of the N:M:P sampling patterns, see http://reality.sgi.com/cpirazzi/lg/packings/

[Dr Mordrid]

For NTSC YCrCb the terms are defined as;

Y = luma
Cr = R-Y
Cb = B-Y

Dr. Mordrid

[Michel Carleer]

Well, there is no difference in the sense that a 4:2:2 means chroma undersampled by a factor of two for PAL as well as for NTSC. There is a difference though in how the chroma information is coded. For PAL, the two chroma signals added to the luminance Y are named U and V. U and V are equal to -(R - Y) and -(B - Y). For NTSC, the two chroma are Cr and Cb, but I don't remember what their exact definition is. Now you know I am a PAL user
The NTSC format should be called YCrCb, not YUV which is strictly PAL. Actually, it is called YCrCb by the video guys, but we PC users like to mess things up a bit.
Someone from the NTSC world could help here?

Michka


[This message has been edited by Michel Carleer (edited 05 October 2000).]