720X480 Vs. 352x480 Difference or no difference?

This has been discussed before but I've never seen and conclusions. Hopefully we'll be able to get one from this thread. The basic argument is that capturing VHS at 720x480 provides no benefit because the resolution of VHS is more in line with 352x480 or even VCD resolution.

Have to shut off comp, ...to be continued :P

Not this subject again!

You should ALWAYS capture at a higher resolution than your final output.

Originally Posted by jagabo

Not this subject again!

But I got pictures. :P

IF you are ever going to convert the footage to another format in the future ie .mp4, DivX etc. you will get better results from the larger 720x480 frame size (at an appropriate bitrate of course)

I'll always cap at 720x480, I have no choice I use a Canopus besides the fact I don't want to lock myself into a lower resolution which brings me to the point of my little experiment. The golden rule is you don't want to scale video up, bearing that in mind why would that not apply to a player scaling as well and if we must scale wouldn't a higher resolution video scale better.

Anyhow here's the specs for both of the following examples, the original source was VHS capped as DV-AVI. From there I madd a 720x480 8000kbps and a 352x480 4000kbps. I don't have a source for directly making 352x480 caps but I don't see why just encoding from a DV-AVI will make any difference.

Anyhow here's a comaprison that was scaled using PowerDVD ver. 4 (note that it's cropped), the full screen capture was 1280x960 the full screen captures and the video they came from are available in the following post.


The 352 is on the bottom:



So i thought aha! It does make a difference so I ran another test with a much better video.


352 on the left:



That really doesn't have much difference at all except for a little bit of the red bleeding into the black so I'm wondering if maybe it's just the player. It is quite old and I appreciate it someone tried the first video.

Files from the first test:
352x480 Screencap , TIF file

720x480 Screecap , TIF file


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


352x480 MPEG

720x480 MPEG

The thing that seems to get lost again and again in this discussion is the distinct difference between a digital sample and an analog line of resolution. They are not the same, and you need a greater digital resolution than the specified analog lines of resolution to capture maximum information.

Consider a 4.2 MHz analog broadcast video bandwidth. By definition, this means that at most 4.2 million cycles per second can be broadcast. That's equivalent to 8.4 million vertical lines per second (on the first half of the cycle, you get white, on the second half, you get black...two lines). When you divide this by 59.94 fields per second, and then by 262.5 scan lines per field, you get 533 vertical lines per scan line. About 16% of each line is blanking, leaving about 84% for active video. 84% of 533 gives about 456 vertical lines.

-------------
Note that this is not the way that "lines of resolution" is quoted.

First, the lines I'm referring to are obviously vertical lines. But the resolution being specified is in the horizontal direction. So what I'm calling vertical lines are normally referred to as "lines of horizontal resolution."

Also, "lines of resolution" is always quoted for a square display. This is to remove the effect of aspect ratio. Since analog broadcast video is 4:3, this means that to get a square display, you'd multiply this value by 3/4. So that means that a 4.2 MHz analog broadcast has approximately 342 lines of horizontal resolution.
-------------

So to capture 456 vertical lines, you only need 456 pixels, right? Not really. Consider that specific situation. If you have 456 alternating black and white lines, and you sample on each peak, then certainly 456 samples will work perfectly (please forgive my very poor sine wave):

Code:

|   _v_         _v_
|  /   \       /   \
| /     \     /     \
|/       \___/       \___/
|          ^           ^

The v and ^ represent sample points. Here you get black and white pixels from your black and white lines, just as you'd want.

But this is analog video. What if that first cycle starts 1/4 of a cycle later (90 degrees out of phase)? Now those same 456 samples aren't aligned so nicely to the vertical lines:

Code:

| v ___       v ___
|  /   \       /   \
| /     \     /     \
|/       \___/       \___/
|       ^           ^

In case you can't tell, every sample point is at the zero crossover. The resulting video will be all gray. You've lost all the lines.

Increasing the sample rate is the only way to ensure that you can encode this second set of lines. Doubling the sample rate will certainly work. But it turns out you don't quite need to go that far.

The ratio of the highest frequency that can be represented by a given digital resolution to that sample rate is known as the Extended Kell Factor. Unfortunately, the exact value of the Extended Kell Factor is somewhat debatable. This is because it depends on your choice as to how much contrast represents a line. In the above example, if you increase the sample rate by just a bit, you'll go in and out of phase with the source sine wave. You'll get varying amounts of contrast from two successive samples. When you are aligned as in the upper image, you'll get great contrast. When you are closer to the bottom one, you'll get less. This variance is an additional, low frequency component we call aliasing. The question is now much aliasing is allowed, and how much variation can you take. This page has some sample images for illustration.

However, in general, something on the order of sqrt(2)/2, or about 70% is a bare minimum for the Extended Kell Factor. So that means to represent our 456 vertical lines, we'll need about 651 pixels.

---------------
By way of confirmation, consider that most DVD players quote "500 lines of resolution." They certainly aren't talking about horizontal lines (vertical resolution), where 480 would be the maximum. They're talking vertical lines (horizontal resolution). And they're talking across the whole screen. So using our same math, 500 lines would require about 714 pixels. (Incidently, while DVDs can have 720 pixels, at the ITU-R BT.601 specified 13.5 MHz sample rate used, only about 714 pixels can fit between the syncs.)
---------------

For VHS tapes, which have about a 3 MHz video bandwidth, doing the same set of calculations: 3 MHz => 6 Mlines => 381 lines/scanline => 320 active lines/scanline => 457 active pixels/scanline. So while 480 is probably enough, 352 probably isn't.

Xesdeeni

I always do my VHS to DVD conversions in full D1 (720x480 NTSC). There is a small but noticeable difference in playback sharpness and detail, especially on large screen HDTV's.

Half D1 might be preferable for a very noisy or grainy VHS, however. Capturing at the lower frame resolution provides a bit of extra noise suppression.

Gshelley61 may be able to comment further on this but it seems to me several years ago, I was trying to duplicate his captures of THX test patterns using Cineplayer while he was using PowerDVD. We could not get results that matched. Lordsmurf and Vhelp suggested using VirtualDubMod to extract stills. We could get consistent matching results then.

If you use a THX test pattern like this multi-purpose pattern, it may easier to compare results. I have recorded the pattern to a VCR for that purpose. Here is the original THX pattern,

Originally Posted by trhouse

Gshelley61 may be able to comment further on this but it seems to me several years ago, I was trying to duplicate his captures of THX test patterns using Cineplayer while he was using PowerDVD. We could not get results that matched. Lordsmurf and Vhelp suggested using VirtualDubMod to extract stills. We could get consistent matching results then.

If you use a THX test pattern like this multi-purpose pattern, it may easier to compare results. I have recorded the pattern to a VCR for that purpose. Here is the original THX pattern,

The main problem with that test pattern for this subject is lack of diagonal references. 352 encoding doubles the effective pixel width causing stepping on diagonals. VHS is softer so it will be less a problem than with sharp sources. The other issues are motion effects and blocking at various bitrates with moving video.

Expect VHS frequency response to be in this range
(picture shows DVD top and VHS bottom).

Originally Posted by davideck

YMMV depending upon the level of detail recorded on tape and the quality of the capture setup (VCR, capture device, display device, etc.).

Do VHS sources contain enough detail to see a difference? VHS tapes recorded and played back on a marginal VCR may not. VHS sources subjected to a lot of filtering and/or Noise Reduction may not. Commercial quality VHS tapes played back on a high quality S-VHS VCR definitely do. I see a difference on my VHS tapes with my equipment.

Sampling Video at 352 is equivalent to a low pass filter somewhere below 240 lines of horizontal resolution. Higher frequency noise and detail will be reduced. The difference between 720 and 352 sampling is even more noticeable with higher resolution sources such as broadcast, S-VHS, Hi-8, Mini-DV, etc.

I would agree wholeheartedly. The 3 MHz bandwidth is indeed a best-case. If the player is old (aren't they all now?), or the tape has been played a lot, or even if it's recorded on LP, it probably doesn't have the full 3 MHz of bandwidth.

And yes, recording with 352 could give an effective low-pass filter. But be careful. When you sample, you must have an analog filter (explicit or implicit) on the input to the sampling circuit to avoid aliasing. Most cheap capture cards have only a hardwired filter frequency. 352 may just pixel drop from a 720/704 filtered capture, instead of filtering for 352. If so, as mentioned above, you are best off capturing at 720 and scaling in software afterwards, so you can have an appropriate filter.

Xesdeeni

Originally Posted by Xesdeeni

...
And yes, recording with 352 could give an effective low-pass filter. But be careful. When you sample, you must have an analog filter (explicit or implicit) on the input to the sampling circuit to avoid aliasing. Most cheap capture cards have only a hardwired filter frequency. 352 may just pixel drop from a 720/704 filtered capture, instead of filtering for 352. If so, as mentioned above, you are best off capturing at 720 and scaling in software afterwards, so you can have an appropriate filter.

Xesdeeni

Most consumer capture cards oversample analog at 27-54MHz to avoid the need for expensive low pass analog filters. 720, 704, or 352 are downsampled with simple digital filtering. This keeps the cost down but allows capture of high frequency noise out to 27MHz that would have been filtered out with a classical low pass filter at 6MHz.

VHS could benefit from a sharp 3MHz low pass luminance filter to keep the chroma subcarrier crosstalk out of the capture. Cost is the issue.

As it stands VHS chroma crosstalk noise into luminance is substantial.

So the digital filtering in the consumer capture cards does not make a lowpass before downsampling? Is there any analogue low-pass luminance filter in the market that would work to connect between the VCR and the capturing card?

Or can this be done in the computer by adding a software filter?
I thought since VHS max bandwidth is 3 MHz there should not be any noise left above 3 MHz? Or do you mean noise from other sources that get mixed into the capture when oversampling?

Which frequency does the croma subcarrier have? Does it help capturing with S-video connection from a VCR with S-VIDEO output?

Originally Posted by ronnylov

So the digital filtering in the consumer capture cards does not make a lowpass before downsampling?

The digital low pass filtering is internal to the chipset. Point is the noise above Nyquist frequency is not filtered out before A/D.

from wikipedia http://en.wikipedia.org/wiki/Analog-to-digital_converter
"Usually, signals are sampled at the minimum rate required, for economy, with the result that the quantization noise introduced is white noise spread over the whole pass band of the converter. If a signal is sampled at a rate much higher than the Nyquist frequency and then digitally filtered to limit it to the signal bandwidth, the signal-to-noise ratio due to quantization noise will be higher than if the whole available band had been used."

In other words there is a tradeoff of noise for cost.

Originally Posted by ronnylov

Is there any analogue low-pass luminance filter in the market that would work to connect between the VCR and the capturing card?

These analog low pass filters are expensive on the order of $100 per component. At retail that could add ~$1000 to the cost of the card. That partly explains the cost difference for broadcast quality capture cards vs. consumer.

Originally Posted by ronnylov

Or can this be done in the computer by adding a software filter?

In theory yes. Software filtering can get CPU intensive.

Originally Posted by ronnylov

I thought since VHS max bandwidth is 3 MHz there should not be any noise left above 3 MHz? Or do you mean noise from other sources that get mixed into the capture when oversampling?

A VCR is full of chroma noise just above 3MHz. It is worse for NTSC where the chroma subcarrier is at 3.58MHz vs 4.43MHz for PAL.

Originally Posted by ronnylov

Which frequency does the croma subcarrier have?

3.58MHz for NTSC. 4.43MHz for PAL. Chroma modulation uses frequencies +/-0.5MHz or more around the subcarrier.

Originally Posted by ronnylov

Does it help capturing with S-video connection from a VCR with S-VIDEO output?

Separation of Y path from chroma can help control chroma crosstalk into the luminance and visa versa. The result is much more effective with S-VHS or Hi8 where luminance overlaps chroma and extends out to 5MHz. It is best to separate luminance and chroma into parallel paths.

Thanks edDV!

I do only make PAL captures and I use a JVC SVHS VCR with S-Video connections so I think this setup is fine because I do not want to spend much money on analogue filters in the capture setup. When I have noisy captures I clean it with some avisynth filters like fft3d which produce a nice and clean output.

In my experience the difference between 352x576 and 704x576 captures of VHS is quite small. Mostly it depends how good the player "upscales" this MPEG-2 resolution at playback. My old cheap DVD player looked "pixelated" when playing 352x576 but with my somewhat newer Pioneer DVD player I can't tell the difference when watched on TV.

However there are differences between capture devices.
See some examples here:
http://www.doom9.org/index.html?/capture/sizes_advanced.html
I have a SAA7134 chip in my capture card and that may explain the small difference between 352x576 and 704x576 when capturing VHS. However normally I capture VHS at 704x576 and make the DVD also at 704x576 to avoid any rescaling problems at playback because I want it to look good on all players.

Originally Posted by ronnylov

In my experience the difference between 352x576 and 704x576 captures of VHS is quite small. Mostly it depends how good the player "upscales" this MPEG-2 resolution at playback. My old cheap DVD player looked "pixelated" when playing 352x576 but with my somewhat newer Pioneer DVD player I can't tell the difference when watched on TV.

These consumer capture cards don't "capture" at 720, 704 or 352. They digitize the analog input at high sample rates with high low pass filtering and then digital filter and downsample to 720, 704 or 352 in the digital domain depending on your capture settings. 720 and 704 use 13.5 Mhz downsample rate for luminance. This ideally passes the equivalent of ~5.75MHz analog (~460 lines of horizontal resolution). 352 roughly passes the equivalent of ~2.8MHz (~230 lines of resolution). Ideally VHS would be sampled midway between between these rates but that option is not available for DVD.

The VHS player is analog. It doesn't "upsample" playback. The player will demodulate luminance (recorded as FM on the tape) and upconvert 500Khz recorded chroma to the subcarrier (4.42MHz for PAL). The difference between cheap and quality VHS players mainly relate to noise and timebase stability performance.

Amen! Using my ATI AIW 8500DV to capture VHS at 352x480 works like a charm with great looking picture that matches the original VHS. Good capture card DOES make a difference.

Relayerman

Originally Posted by lordsmurf

By the way, you CANNOT use compressed video data to make comparisons between sharpness on resolutions. Not all frames are alike, and your master frame on one may not match the master frame on another.

You'd have to use an uncompressed frame.

If you are refering to my sentance

"720 and 704 use 13.5 Mhz downsample rate for luminance. This ideally passes the equivalent of ~5.75MHz analog (~460 lines of horizontal resolution). 352 roughly passes the equivalent of ~2.8MHz (~230 lines of resolution). Ideally VHS would be sampled midway between between these rates but that option is not available for DVD."

I did say "ideally" and yes this would apply to perfect A/D and filtering which is never the case and to uncompressed capture. I was only using those figures to relate 720/704 and 352 sampling to analog video.

Capture, resampling and MPeg2 compression issues are three separate subjects that often get confused.

I would argue that 352x480/576 is near adequate for VHS but I'd rather use 640x480/576 or 704x480/576 if quality was was a major concern (e.g. home camcorder videos or direct dubs from a high quality source).

I don't agree that 352x480/576 is adequate for broadcast NTSC, PAL, S-Video, Hi8 or
laserdisc. You can use it for data reduction but quality will suffer in the compromise.

Broadcast NTSC is a special case. If a quality comb filter is used to decode NTSC, then the capture will have detail out to 4.2 MHz. If not 352x480 may be adequate but not ideal.

Originally Posted by lordsmurf

Originally Posted by edDV

I don't agree that 352x480/576 is adequate for ... S-Video, Hi8 or laserdisc. You can use it for data reduction but quality will suffer in the compromise.

Do you mean S-VHS, not s-video? If so, I agree on that one. Same for Hi8 or Laserdisc. There is more data there than 352x480, although it's still quite a bit under 720x480. Here you'd have to just use best judgment and make decisions on secondary factors.

Sorry, yes S-VHS. I didn't proof read that one.

The only choice they give us is 720/704 or 352. Media is cheap.

In short terms:
If your source is VHS/ SVHS then capture the higher you can, filter, encode to half D1.

Few cards have good enough capture abilities on the 352x480/576 framesize. So capture full D1 just to be sure that you capture everything (especially when you are a newbie).