Capture Cards

[Torrential]

Hello All,

I am looking into buying a capture card, and i've narrowed it down to two choices: Canopus or an ATI AiW. I'm leaning more towards Canopus. What is the best Canopus Card (max I can spend is $250, MAYBE $300).

Here is what I need:

.AVI Capture at 640x480 with Very High Quality
PCI Card

Here are the specs of my stuff:

Dish Network Sattellite Receiver
Windows XP Pro
1GB RAM
2.8GHz P4

I have the ATI TV Wonder Pro Card and it just doesn't produce enough quality for me to like, It must meet my criteria that I specified, the final output will be an XviD/DivX AVI File.

[Torrential]

Opinions from anyone?????????????????

[redwudz]

I use the ADVC-100, no complaints. Best feature is low CPU load and locked Audio/Video sync. Easy to edit also. I occasionally feed the DV to Mainconcept encoder for MPEG on the fly. Not as good of quality with this, but quick and easy for TV quality video. I like the external converter as I can switch it between computers and it's usable with different platforms such a my Mac.

I have some experience with older ATI cards and though I liked the quality, the CPU load was high and to get quality captures it took a lot of HD space. Hardware MPEG encoding would probably be better, but then editing can be a problem.

[Torrential]

With the one you have, you can choose to capture in a regular AVI, not DV-AVI, right?

[redwudz]

No, the only output to the computer is DV. The codec is set by the hardware in the ADVC and the video is compressed there. It uses about 13GB of hard drive space for each hour of video.

[Torrential]

Are DV-AVI files editable in VirtualDub and can they be compressed with XViD?

[redwudz]

It's easy to edit DV. Most editors will accept it. With VirtualDub, you just need to add a DV codec so VD will recognize the format. I use the Panasonic DV codec.

Then you can encode with VD in XVID or DIVX format, or other AVI type formats. You can also frameserve the VD output to a MPEG encoder like TMPGEnc and you don't have to create an intermediate file.

VirtualDub will not 'capture' DV directly. I use WinDV to transfer the DV via Firewire to the HD.

[Torrential]

Thanks for all the info -- just one more question to clear things up. What is the difference between DV and regular AVI? Is there a big difference in quality?

[lordsmurf]

Yes. DV in NTSC has colorspace compression (4:1:1) in addition to filesize compression, about 5:1 ratio. For shooting new video, it's fine. For converting old VHS tapes, there are issues.

The ATI AIW card you had looked at can do any number of AVI codecs as well as MPEG. As far as a capture card goes, they're hard to beat in terms of quality and functionality.

[junkmalle]

Thanks for all the info -- just one more question to clear things up. What is the difference between DV and regular AVI? Is there a big difference in quality?

That's like asking "Are oranges better than regular boxes?" There's no such thing as "regular AVI". AVI is a container. You put audio and video inside it. The audio and video can be uncompresses or compressed with any number of COmpression/DECompression (codec) schemes. DV is one codec used in AVI containers, and the native codec of DV camcorders.

DV is mildly compressed and easy for editing programs to deal with. It caputures the color portion of the picture at 1/2 the resolution (both dimensions) of the black and white portion. This is similar to movie DVDs (and human perception), but a little less than a native video signal. DV AVI files are about 13 GB/hr. DV capture devices perform the DV compression inside the capture box, then send the digital data to the computer via firewire. (Some DV boxes are actually cards that plug into a PCI slot, but they only use the slot for power, they still send the compressed DV data to the computer via firewire.)

Cards that capture something closer to the raw video signal capture color at half the resolution horizonally, but at full resolution vertically. If you save this uncompressed it comes out to about 75 GB/hr. Lossless compression (HuffYUV for example) can reduce that by roughly 1/2. Most capture programs allow you to select different codecs for realtime compression (to DVD compatible MPEG for example) by the computer while capturing. But since the computer must perform the compression in real time it can't always do the best possible job, and you may need a fast computer for the more sophisticated codecs (for more compression). So saving in uncompressed (or losslessly compressed) format, and compressing the video later (when the computer can take its time) will usually give you a better result.

A third major type of capture device captures directly to MPEG format. These capture the video signal and convert to MPEG internally, then send the MPEG data to the computer (they can be external boxes or internal cards). These can be the most bulletproof at capturing (ie, without dropping frames) and the devices can capture with DVD compatible MPEG settings -- alleviating the need for time-consuming conversion later for DVD. Full 720x480 MPEG captures can run anywhere from about 2 to 8 GB/hr. But MPEG is not as easily edited as DV or the uncompressed formats. You will always get some MPEG artifacting (whether this is significant or not is in the eye of the beholder). And it usually has half the color resolution, like DV.

[Torrential]

So it looks like this device would not be good for capturing TV, does anyone have a sample of a TV capture with this device?

[junkmalle]

Canopus DV devices should be fine for satellite TV caps. Dish network is all MPEG compressed so it already has the color portion of the picture at half the resolution of the black and white. You might get very slightly better results with an ATI AIW or another raw YUV capture card but you'd probably only be able to see any difference with test images created specifically to show them. And capturing DV is much less prone to problems in my experience.

[lordsmurf]

The way I describe DV footage versus uncompressed ... is the same way I describe real life against CG recreations .... it just doesn't look right. Something is different, and most people will notice it, though not everybody will say something. But it's not really noticed until it's entering a second re-compression phase, which is what DV transfer of standard signals is (as opposed to original signals created in a camera, coming from uncompressed real life).

And actually, in terms of colorspace compression, MPEG for DVD is less compressed than DV is. At least in terms of human visual perceptions. And then, only NTSC. PAL DV is 4:2:0 (and if therefore quite decent to work with), whereas whoever chose 4:1:1 for NTSC should be flogged with a VCR for stupid choices. Or at very least, have created DV converters that operate 4:2:2 or 4:2:0 NTSC without requiring NLE cards (some advanced DV NTSC cards are actually 4:2:2 native).

It all depends on what you want to do. DV devices are an easy way out, in terms of hardware/software setup. There's nothing to do. Though it does require MPEG encoding phases to get onto DVD. And contrary to stupid salesmen, it's JUST a transfer device, there is not additional magic in the box (TBC, improve image, etc ... not gonna happen).

I opted for the ATI AIW because it gave me either uncompressed (or codec of my choice) AVI ability, and then MPEG abilities too. I'm no idiot, so installing the hardware and software was not even remotely difficult. And I'm using a decent computer (and have common sense to not operate a bloated system) so dropped frames and audio issues never happen.

You have to weigh the good and bad and see which is the lesser of evils. Both of the cards in question operate great, and give the results as described. DV in one case, AVI/MPEG in the other. Which do you prefer?

[Fos]

I recently have bought an Ati 9200. The mainly problem is MV effect. So please, dont buy it at least you have a tbc.
I want to buy another card , i cant affroid a tbc, and i dont know where i can buy it. I will buy a pinnacle dc2000 or 1000 (if i find someday in ebay) or the canopus advc 50. Possibily the canopus , because the canopus dv codec is the best!

[junkmalle]

I've never seen clear examples of what effect color subsampling has on images. So I made an RGB test image and converted it to NTSC DV (Canopus codec) and MPEG2 with Ulead Media Studio Pro 7. The MPEG bitrate was 8000 kbps and the color subsampling was 4:2:0. I'm not sure what MPEG2 codec Ulead includes but I also tried TMPGEnc and it gave very similar results. I used progressive settings for all conversions. Here are some crops, enlarged 4x with a nearest neighbor filter so you can see indivudual pixels:



The left column is from the original uncompressed RGB AVI file. The original cyan blocks are 8x8 pixels. As you move down the column the block moves 1 pixel to the right and 17 pixels down. This is to provide different alignments of the luma and chroma channels.

The middle column is 4:1:1 NTSC DV. As you can see the colors are smeared horizontally but not vertically. The top and bottom blocks happen to have the luma and chroma samples perfectly aligned so the original 8x8 block is perfectly recreated. This is of course an aberation of the test image. You would rarely see this happen in any real world source.

The right column is 4:2:0 NTSC MPEG2. Colors are smeared both horizontally and vertically, but only half as much horizontally as the DV sample (2 vs 4 pixels). Exactly what one would expect. I, B and P frames all looked similar.

Some things to keep in mind:

1) These are pure digital conversions from a digital RGB source. At no point have any of these been through an analog stage. You won't find any analog source as clear and clean as this RGB video.

2) I originally used blocks of different colors and different sizes. All showed similar effects but some colors smeared more than others. The 8x8 cyan blocks in these samples are typical. Grey blocks did not smear at all.

3) Satellite captures (via s-video cable) will be from an analog source, compressed with MPEG for transmission, decoded back to analog video for display, and converted again to digital by whatever capture device you use. All these conversions will only make things worse...

[hrlslcbr]

I have the Theatrix (the same as the TV wonder ELITE) and it has excellent quality. You can capture to MPEG-2 in hardware or to uncompressed AVI (and use any codec you want in your capturing programme).

EDIT: junkmalle: I tried your 'test' with the newest build of x264 and none of the squares are like the source (all of them are blurry). Interesting.

[edDV]

I've never seen clear examples of what effect color subsampling has on images. So I made an RGB test image and converted it to NTSC DV (Canopus codec) and MPEG2 with Ulead Media Studio Pro 7. The MPEG bitrate was 8000 kbps and the color subsampling was 4:2:0. I'm not sure what MPEG2 codec Ulead includes but I also tried TMPGEnc and it gave very similar results. I used progressive settings for all conversions. Here are some crops, enlarged 4x with a nearest neighbor filter so you can see indivudual pixels:

image

The left column is from the original uncompressed RGB AVI file. The original cyan blocks are 8x8 pixels. As you move down the column the block moves 1 pixel to the right and 17 pixels down. This is to provide different alignments of the luma and chroma channels.

The middle column is 4:1:1 NTSC DV. As you can see the colors are smeared horizontally but not vertically. The top and bottom blocks happen to have the luma and chroma samples perfectly aligned so the original 8x8 block is perfectly recreated. This is of course an aberation of the test image. You would rarely see this happen in any real world source.

The right column is 4:2:0 NTSC MPEG2. Colors are smeared both horizontally and vertically, but only half as much horizontally as the DV sample (2 vs 4 pixels). Exactly what one would expect. I, B and P frames all looked similar.

Some things to keep in mind:

1) These are pure digital conversions from a digital RGB source. At no point have any of these been through an analog stage. You won't find any analog source as clear and clean as this RGB video.

2) I originally used blocks of different colors and different sizes. All showed similar effects but some colors smeared more than others. The 8x8 cyan blocks in these samples are typical. Grey blocks did not smear at all.

3) Satellite captures (via s-video cable) will be from an analog source, compressed with MPEG for transmission, decoded back to analog video for display, and converted again to digital by whatever capture device you use. All these conversions will only make things worse...

While this is true for full bandwidth color sampling, the luminance resolution is not affected so long as you avoid RGB transforms and keep luminance and chroma separate. 4:1:1 (13.5MHz luma, 3.475MHz chroma) oversamples all of the formats below so your experiment vastly overstates the issue.

Typical home analog sources are all variations of Y U V as follows (worst to best).

VHS, 8mm videotape -- max effective capture raster (288x480or 288x576)
(Luminance 3MHz ~240 lines of H resolution, UV less than 0.5MHz ~ 40 lines of H resolution)

NTSC TV tuner -- max effective capture raster (352x480)
(Luminance 4.2MHz ~336 lines of H resolution, IQ @1.0MHz ~80 lines)

PAL TV tuner -- max effective capture raster (480x576)
(Luminance 5MHz ~400 lines of H resolution, UV @1.2MHz ~100 lines)

SVHS, Hi8 -- max effective capture raster (480x480 or 480x576)
(Luminance 5.1MHz ~410 lines of H resolution, less than 0.5MHz ~ 40 lines of H resolution)

MiniDV and Digital8 -- max effective capture raster (720x480 or 720x576)
(Luminance 6.75MHz ~540 lines of H resolution, UV 1.7MHz ~ 135 (280 PAL) lines of H resolution)

4:1:1 offers more than adequate sampling of all the formats above. Nyquest bandwidth is 6.75MHz luma and 1.7MHz chroma. You would get no better results sampling at 4:2:2 or even 4:4:4. The source chroma bandwidth is not there.

4:1:1 can also be used as a production format. The chroma pixels line up with the first luminance sample. Y, U and V maintain spacial resolution during transformation. 4:2:0, like MPeg2 is a 2D display resolution and is not intended for further processing.

If you consider the bandwidth difference between Y and UV in all instances above, you will begin to understand the damage to luminance resolution that results when a RGB conversion is made. The conversion will dilute the resolution of luminance.

R = Y + (1.4075 * (V - 128));
G = Y - (0.3455 * (U - 128) - (0.7169 * (V - 128));
B = Y + (1.7790 * (U - 128);
Y = R * .299 + G * .587 + B * .114;
U = R * -.169 + G * -.332 + B * .500 + 128.;
V = R * .500 + G * -.419 + B * -.0813 + 128.;

[edDV]

The way I describe DV footage versus uncompressed ... is the same way I describe real life against CG recreations .... it just doesn't look right. Something is different, and most people will notice it, though not everybody will say something. But it's not really noticed until it's entering a second re-compression phase, which is what DV transfer of standard signals is (as opposed to original signals created in a camera, coming from uncompressed real life).

Whatever you are seeing it isn't due to 4:1:1 sampling. Almost everything you see coming from local TV, cable and network news in most parts of the world is 4:1:1 DV, DVCAM or DVCPro that has been heavily edited.

4:2:2 (Digital Betacam) is typcally used only for higher end production projects or network program time shifting.

http://www.adamwilt.com/DV-FAQ-tech.html#colorSampling

And actually, in terms of colorspace compression, MPEG for DVD is less compressed than DV is. At least in terms of human visual perceptions. And then, only NTSC. PAL DV is 4:2:0 (and if therefore quite decent to work with), whereas whoever chose 4:1:1 for NTSC should be flogged with a VCR for stupid choices. Or at very least, have created DV converters that operate 4:2:2 or 4:2:0 NTSC without requiring NLE cards (some advanced DV NTSC cards are actually 4:2:2 native).

Higher end workstations do use 4:2:2 SDI I/O and 4:2:2 cards but they are also working with 4:2:2 inputs such as Digital Betacam and 4:2:2 film transfers. This is the world of professional production and >$30,000 camcorders and VCRs. DVD is not 4:2:2 but 4:2:0, half the chroma resolution in both H&V but the eye doesn't care. You can't see even that much resolution in color.

4:1:1 is a prefectly adequate production format for those without high end budgets. 4:1:1 oversamples any source the home user is likely to capture. 4:1:1 has excellent multigeneration performance where 4:2:0 does not. 4:1:1 can be processed with a higher quality end result than 4:2:0. DV capture levels match exactly the DVD MPeg2 format.

I opted for the ATI AIW because it gave me either uncompressed (or codec of my choice) AVI ability, and then MPEG abilities too. I'm no idiot, so installing the hardware and software was not even remotely difficult. And I'm using a decent computer (and have common sense to not operate a bloated system) so dropped frames and audio issues never happen.

You have to weigh the good and bad and see which is the lesser of evils. Both of the cards in question operate great, and give the results as described. DV in one case, AVI/MPEG in the other. Which do you prefer?

For VHS and TV captures, both DV and AIW will work and you are going to have to convince me that capturing at greater than 4:1:1 sampling results in anything but increased noise. I think you are going to choose between these two devices for some other reason than those stated above.

[lordsmurf]

The conversion will dilute the resolution of luminance.

I think this statement alone is the important part.

Going from 4:2:2 (tv/VHS) to 4:1:1 (NTSC version DV) to 4:2:0 (DVD) is the problem, in terms of quality being maintained during the entire process. Or rather, the lack of quality. You notice it most in reds, greens and the contrast values.

The loss is most definitely in the processing, not something else. Process of elimination.

[edDV]

The conversion will dilute the resolution of luminance.

I think this statement alone is the important part.

Going from 4:2:2 (tv/VHS) to 4:1:1 (NTSC version DV) to 4:2:0 (DVD) is the problem, in terms of quality being maintained during the entire process. Or rather, the lack of quality. You notice it most in reds, greens and the contrast values.

The loss is most definitely in the processing, not something else. Process of elimination.

The "conversion" to be avoided was YUV to RGB.

First the first 4 is luminance and remains the same from capture to DVD (unless you choose to process it). All resolution detail and "contrast" is contained in luminance.

The TV we are talking about here is analog (from tuner or set top box) NTSC has maximum chroma bandwith of 1 MHz for I and Q in modern equipment.* Neither TV nor VHS are 4:2:2. Let's calculate the equivalent.

4:2:2 is (Nyquest bandwidth) 6.75 3.375 3.375 MHz
4:1:1 is (Nyquest bandwidth) 6.75 1.690 1.690 MHz

NTSC TV (best case bandwidth) is 4.2 1.0 1.0 MHz or 2.5 0.6 0.6
VHS 8mm (best case bandwidth) is 3.0 0.5 0.5 MHz or 1.8 0.3 0.3
Betacam SP (best case bandwidth) 4.5 1.5 1.5 MHz or 2.7 0.9 0.9

So as you can see, 4:1:1 oversamples even Betacam SP.

DV format is captured directly to 4:1:1 YUV. The MPeg2 encoder will interpolate 4:1:1 into 4:2:0. This is done for U and V only and has no reported color errors that I have heard. Although the UV pixels are relocated, the chroma bandwidth is still beyond human perception at 1x picture size.

*Original NTSC had chroma bandwidth at 1.2MHz for I and at 600KHz for Q. Crosstalk problems prevented the wider I bandwidth from being used in practice. Typical TV sets and VCRs process I and Q at 500kHz or less.

[hrlslcbr]

edDV: I always understood that if you capture to 4:1:1 and then encode to a format that uses 4:2:0, you would en up with '4:1:0' REAL colour samples and the rest would be interpolated at some point. Isn't that right?

[edDV]

edDV: I always understood that if you capture to 4:1:1 and then encode to a format that uses 4:2:0, you would en up with '4:1:0' REAL colour samples and the rest would be interpolated at some point. Isn't that right?

Well it depends how the encoder works.

4:2:0 has no "real" chroma pixels, they are all repositioned relative to Y. For 4:1:1 to 4:2:0 you would interpolate from high resolution vertical and lower resolution horizontal. 4:1:0 would be something very different whatever it is.

4:1:1 looks like this (chroma only 4 lines)

uv___uv___uv___uv
uv___uv___uv___uv
uv___uv___uv___uv
uv___uv___uv___uv

4:2:0 looks something like this (chroma only 4 lines)

Co-sited, *=interpolated if converted from 4:1:1

u-_u*_u-_u*_u-_u*_u-_u*_
-v_*v_-v_*v_-v_*v_-v_*v_
u-_u*_u-_u*_u-_u*_u-_u*_
-v_*v_-v_*v_-v_*v_-v_*v_

or this (all interpolated)

_u_v_u_v_u_v_u_v_
_v_u_v_u_v_u_v_u_
_u_v_u_v_u_v_u_v_
_v_u_v_u_v_u_v_u_

http://www.adamwilt.com/pix-sampling.html

To convert 4:1:1 to 4:2:0 interpolation would be used. Keep in mind that the eye would see all this as a chroma blur and probably not see any difference between the two.

Also remember, there are two kinds of 4:2:0. One has half the U or V pixels co-sited with Y and the other has all chroma pixels interpolated.

[lordsmurf]

To convert 4:1:1 to 4:2:0 interpolation would be used. Keep in mind that the eye would see all this as a chroma blur and probably not see any difference between the two.

But that's my point. Many people CAN see this, in terms of red/green and varying contrast artifacts.

And then add in the fact some of the codecs are rather poor (including the self-proclaimed "divine" Canopus codecs), and that's where the true problem comes in.

I'll agree with you that, in theory, it should not be too different. But the codecs, and maybe even the encoders (as you mentioned, but less likely), are the villians that make DV 4:1:1 to DVD 4:2:0 a less savory option than methods that involve less colorspace changes.

It's all the nature of the tech. These methods work and look decent. But for the discriminating eye, there are better. And sadly, better often costs the same or less, so why blow $$$ on an expensive solution that is not as solid a method?

I'll go ahead and answer my own question: user error. Or rather, many of these DV devices prevent any form of user error, and are easy. That's really the only thing they have going for them. It's decent on quality and easy to operate. Better cards usually take a little more work, and often have a slight learning curve involved. There's really no mystery to this, and there's really no need for Canopus to blow smoke at people with "best quality ever" marketing, they just need to hit the "it's super easy" audience, because that's their typical buyer.

[Torrential]

redwudz, can you post a sample capture (if you can - compress it with something so its not huge). Thanks!

[edDV]

To convert 4:1:1 to 4:2:0 interpolation would be used. Keep in mind that the eye would see all this as a chroma blur and probably not see any difference between the two.

But that's my point. Many people CAN see this, in terms of red/green and varying contrast artifacts.

And then add in the fact some of the codecs are rather poor (including the self-proclaimed "divine" Canopus codecs), and that's where the true problem comes in.

I don't know about Canopus DV codecs being any better than others. We are just talking DV format here. DV format has greater sampling bandwidth than any format up to Betacam SP. The differences of 4:1:1 vs 4:2:0 are mainly of interest to software writers. The human eye cannot resolve the different geometric layouts at 1x. None of this affects luminance where the eye is resolution sensitive.

If you zoom into the picture some minor differences can be shown with test patterns and graphics but not as much for camera generated video. 4:1:1 has been shown to have superior multi generation performance because the chroma pixels stay fixed in X-Y space vs 4:2:0 where interpolated chroma pixels get reprocessed each pass.

In terms of other home video problems,this one is third order.

I'll agree with you that, in theory, it should not be too different. But the codecs, and maybe even the encoders (as you mentioned, but less likely), are the villians that make DV 4:1:1 to DVD 4:2:0 a less savory option than methods that involve less colorspace changes.

It's all the nature of the tech. These methods work and look decent. But for the discriminating eye, there are better. And sadly, better often costs the same or less, so why blow $$$ on an expensive solution that is not as solid a method?

I'll go ahead and answer my own question: user error. Or rather, many of these DV devices prevent any form of user error, and are easy. That's really the only thing they have going for them. It's decent on quality and easy to operate. Better cards usually take a little more work, and often have a slight learning curve involved. There's really no mystery to this, and there's really no need for Canopus to blow smoke at people with "best quality ever" marketing, they just need to hit the "it's super easy" audience, because that's their typical buyer.

I just don't see a problem especially for relatively higher bandwidth camcorder and TV capture. I'm open to theroies about low bandwidth VHS capture but I doubt there is a practical problem at all.

[lordsmurf]

It is more noticeable on laserdisc and VHS source.