Resolution size?

[lordsmurf]

You won't like the way I'm gonna answer you anyway....

8)

[Wilbert]

Just when I finished saying that many don't like that practical determination, a doom9 forum member jumps, by linking us to doom9....

I see that Xesdeeni already explained to you why this 300 vertical resolution is false[*]:

https://www.videohelp.com/forum/archive/t210804.html

"The discrete sampling of the video signal vertically, even a completely analog one, is already done, when the image was divided into scanlines. Even if the video is filtered prior to being converted to scanlines (by the camera or electronically), there are still precisely 480/576 vertical samples. If you downsample vertically, you will lose information."

I take it you don't agree with this.

junkmalle even provided hard emperical evidence: same thread post "2004 Mar 14 15:04".
[*] There's something called perceived vertical resolution. Stuff like Kell factor and Circle Rule comes into play.

http://members.aol.com/ajaynejr/vidres.htm

Quote from that page (from a NTSC example):
"While the TV set may be capable of higher resolution (needed for the best playback of DVD) the most resolution that will be displayed for received TV broadcasts is 330 lines."

So, that's only about the amount of detail (vertically) that you can see during display.

[lordsmurf]

I agree with the x480 assessment, as well as many other things said here. But not all. I approach video from a practical stance. Theory doesn't mean much to me if I cannot see it.

Video theory is a little like religion to me. Lots of stuff to read and talk about, but at some point, things you cannot see are pointless to discuss. I'm also not impressed by math, as it is just as much art as it is science.

[SatStorm]

I don't even understand the english Xesdeeni is speaking....
I'm stupid.

The theory behind this hobby is not at my interests. That's why I'm at a mainstream and not at a technical / theory forum.
I'm only interest to do my job in the best possible and fast way, using practical terms and ways, that won't transform me to a nerd or something. All this theory you bring on topic, won't bring new enthusiasts in the video scene. Makes this hobby (video) looks a nerd thing, not a hobby for the average joe, which trully is...

Also you found very fast that older post.... Maybe you are related somehow with him or you are the same person. I can second that because of your skills at english. Don't mention your NTSC examples...
Funny and strange, for a user from Holland (PAL) like you...

[Wilbert]

I was googling around for "interlace barrier", and that thread popped up. And no, I'm not related to him. Other then that he also posts at doom9 sometimes.

Don't mention your NTSC examples...
Funny and strange, for a user from Holland (PAL) like you...

??? The reference I gave used a NTSC example. Why would I talk about PAL examples then?

The theory behind this hobby is not at my interests. That's why I'm at a mainstream and not at a technical / theory forum.

We are not talking about theory here. Like I said

junkmalle even provided hard emperical evidence: same thread post "2004 Mar 14 15:04".

If that's not practical, then I don't know what practical is. Everyone can repeat that test without knowing anything about theory. I suggest you do it, you may learn something

I don't care if you don't care about theory, but you could at least give correct advices:

- VHS is about 300 x 576 for capping purposes (with bt8x8/ati [*} you should cap higher, because the resizers are not good)
- SVHS is about 500 x 576 for capping purposes

See, you don't need any theory at all, to make this statement.

[*} I guess LordSmurf disagrees with this, but I'm still waiting for the ati (theater) pics

[SatStorm]

I saw a typo I did:

I wrote that "VHS is about 240 x 300, interlace, scretch in a 720 x 576 interlace Canvas".
The correct is that VHS is about 320 x 300, interlace, scretch in a 720 x 576 interlace

~240 x 300 is the LP VHS mode (sorry for this, I'm still on vacations and my mind is still on the beach, looking naked blonde females from Nordic countries.....)

SVHS is about 320 x 400, interlace, scretch in a 720 x 576 interlace canvas

We don't capture the VHS / SVHS info, we can't capture that at all. The closest thing we can do, is to capture the Canvas that include that info. Then we choose: we can reconstruct the VHS or we can emulate the Canvas with the VHS/SVHS info in it.
The second is easier to do.

352 x 576/480 is not true VHS neither true SVHS. It is an overkill for VHS and it is close to SVHS. But for various reasons, that framesize it is the easier and safer thing we can do with those kind of sources.

[lordsmurf]

There is an interlace barrier. I saw it in a spec sheet myself long ago, and others I know have talked about it too. As far as links, I'm pretty sure I saw that offline. It's a tech limit of the format. There was no need to interlace below that, for one. Interlace is only really for tv sets, as it applies to MPEG format. And only a few formats have it, MPEG-2 is one of them, MPEG-1 is not.

(sorry for this, I'm still on vacations and my mind is still on the beach, looking naked blonde females from Nordic countries.....)

Pics? :P

[LSchafroth]

Pics? :P

Yes, what he said... pics???

LS

[iantri]

Wilbert; the problem I have with the idea of capturing at 368x480 is that it is not a DVD-compliant resolution, which, as far as I'm concerned, makes it unreasonably inconvenient to have to resize things to acheive that. I don't see why one would bother with futzing about with odd capture sizes when if you are going to resize anyway, you may as well capture 720x480 (since, depending on capture software it may not be possible to do 368x480)

I also disagree with your statements about the resolution of a TV screen; that figure refers to horizontal lines of resolution; which of course varies from set to set. Horizontal lines of resolutions being how many lines can be resolved, horizontally. Ex:


On a TV set with "330 lines of resolution" you would be able to discern the individual lines in that picture if there were 330 of them.

I think arguing over the exact "resolution" of an analog image is rather pointless; it is going to vary depending on the quality of your capture device, VCR, and source. Of course, the figures quoted are good ballpark figures.

(BTW, for anyone having trouble with the idea that VHS does not capture a full 480 lines horizontally, remember that a frame is made up of two seperate fields; you will have less detail than full resolution in each field and your combined frame will have less than 480 lines of detail, but will have 480 lines in it.)

[junkmalle]

I went ahead and did some experiments. I started with this resolution test pattern from another recent thread:

https://www.videohelp.com/forum/images/guides/p1013792/Resolution.png

This 720x480 pattern has a lot of moire patterns in the higher resolution sections but it is pretty clean in the lower resolution areas we are interested in for VHS.

With Ulead Media Studio Pro 7 I created a DVD compatible MPEG file of that image slowly panning across the screen -- about 1/8 pixel per frame. Of course this means MSP7 had to resample the image for each frame. I burned the file as a movie DVD with Ulead DVD Movie Factory. When watched on my 32 inch CRT based TV connected to a DVD player via component video cables. All the resolutions below 400 lines (I'm talking about the horizontal resolution patterns -- ie vertical lines) were clear and relatively free of moire patterns. Higher resolution sections showed moire patterns and strobing as the image panned across the screen.

I then recorded the DVD output to an standard VHS deck via a composite cable. When played back on a TV connected via composite cable I could clearly see individual lines near the 200 line marker. By 250 the lines had disappeared into a gray blur. Around 225 lines the there was still some evidence of the lines but it was getting blurry and strobed a bit. So I'm calling the resolution of this player, nominally, 225 lines.

Now that test pattern is labeled the way resolution is measured in analog video systems. The "lines" of resolution is measured over a width equal to the height of the picture, not the full width of the picture. That is, it is measured over a 3:3 portion of the image, not the full 4:3 portion. So to get the number of lines across the entire picture we need to multiply by 1.33 (4/3). 225 * 1.33 = 300. As a sanity check, I cropped a small section of the pattern in that area and duplicated it across the screen. The result was 299 lines -- close enough.

OK, now we know the horizontal resolution of a typical VHS deck is around 300 lines across the entire width of the screen. In a later post I will discuss why digitally sampling a 300 line analog image with 300 (or even 352) pixels is insufficient.

[Wilbert]

Wilbert; the problem I have with the idea of capturing at 368x480 is that it is not a DVD-compliant resolution, which, as far as I'm concerned, makes it unreasonably inconvenient to have to resize things to acheive that.

Of course I agree with that. You can also cap at 352x576 and accept quality loss

I also disagree with your statements about the resolution of a TV screen; that figure refers to horizontal lines of resolution

Good that some wants to argue about it, because I see that you are right.

correct is that VHS is about 320 x 300

So, where's is this vertical resolution of 300 coming from?

http://members.aol.com/ajaynejr/kell.htm

I guess it's the kell factor times the number of scanlines, ie about 0.7*480 = 330?

"TIn the NTSC TV picture made up of 480 scan lines (the rest of the 525 lines are "black" and contain format information), applying the Kell factor of 0.7 gives about 335 lines of vertical resolution. This closely matches the broadcast horizontal resolution of 330 lines. This was not an accident; the TV picture was originally designed to have about the same resolution both horizontally and vertically."

We don't capture the VHS / SVHS info, we can't capture that at all. The closest thing we can do, is to capture the Canvas that include that info.

What's a Canvas?

[lordsmurf]

Forget Kell.
You're just going to make it more confusing than it should be.

[iantri]

The idea of there being a vertical resolution of only about 300 pixels and capturing the entire 'canvas', basically means that, while the video DOES technically have 480 lines, VHS tape does not capture that in full detail. However, 480 lines must be captured to get the complete interlaced video from a capture card.

Think of the video in terms of fields and not frames. VHS tape is a series of fields, not combined frames like digital video. Now imagine the fields are blurred. Recombine into frames. You have a 480 line image but 480 individual lines can not be discerned.

[trevlac]

This is one of the best topics.... many people don't know what they are talking about ... and many know quite a bit .... I will try to be nice.

The BT chip does a vertical filter if you cap at/below 360.

I did this test more than a year ago ...
http://trevlac.us/pics/Sharpness360.jpg
http://trevlac.us/pics/Sharpness368.jpg

Notice the duplicated thin horizontal line in the 360 pic. AKA below ~368x480 the BT is a mess. And true .... what's the point at caping at 368?

Max VHS bandwidth is stated as 3MHz. This is really a -4db figure (or 63%) of the original value.

Horizontal TV Lines are the traditional measure of resolution. The rule of thumb (for NTSC anyway) is 80 lines for each 1MHz. So VHS supports ~ 240 HTVL where the high frequencies are actually a bit darker and blurred to only 63% of their original value.

HTVL is a measure across the screen. Up and down (vertical) has the same number of lines regardless of the analog source (for NTSC). 576 is the PAL number, 480-486 is the NTSC number (differences of 6 are because it depends on how you count). VHS / S-VHS / LD / DVD whatever ... all have the same vertical resolution. This is how a TV raster works. If you think differently .... you need to read a little.

Because our screens are 4:3 ... to translate to a 4:3 DVD frame size ... you must multiply by 1.333. So 240 HTVL * 1.3333 = 320x480.

Max for VHS is 320x480 (576) in comparable DVD frame sizes. End of story ...

I'm not saying that a VHS tape has that much detail. They probably don't. However ... as the BT case proves .... there a many ways to screw up the picture .... so you should test and see.

BTW: For the people who like pics ....

Here is that resolution pic: http://trevlac.us/pics/Resolution1.png
Here is what VHS can do to it: http://trevlac.us/pics/VHS712.png

If you look closely ... you will see that the full vertical (up and down) resolution is pretty much there. Horizontal ... you can just make out the 3MHz circle). If you read the HTVL wedges (the ones that run up and down in this case) ... you will see they measure at around 240.

So .... there ya go ... clear explanation ... some pics ... not too much math ...


BTW: I i i say ... SatStorm my boy ... you're as crazy as a junebug in a bottle of my pappy's huckelberry moon shine.

--- If you want I'll say the same thing to Wilbert over at doom9 ---

[trevlac]

The idea of there being a vertical resolution of only about 300 pixels and capturing the entire 'canvas', basically means that, while the video DOES technically have 480 lines, VHS tape does not capture that in full detail. However, 480 lines must be captured to get the complete interlaced video from a capture card.

Think of the video in terms of fields and not frames. VHS tape is a series of fields, not combined frames like digital video. Now imagine the fields are blurred. Recombine into frames. You have a 480 line image but 480 individual lines can not be discerned.

Where did you get this? Let's just talk vertical. Up and down.

A vhs and a broadcast and a LD and a DVD (analog out) signal is a continuous signal that has little sync points that tell the display device when a new line starts and when a new field starts. There are always the same number of lines. The detail in the signal can change, but that effects the horizntal resolution.

To repeat ... All analog source is the same vertical (within a standard like NTSC). You always see the analog ... it follows the standard.

Regardless of kell and all that stuff .... and how you want to measure ... it is always the same. No real point in talking about vertical. I will look for some reference sources for ya.

Might as well just say the number of fields changes from format to format. Sometimes you get 60 sometimes less ....

Here is a detail picture of what is in an NTSC signal:
http://trevlac.us/pics/ntsc.gif

[lordsmurf]

What trevlac said. I think we agree 100% for once.

[iantri]

Well, I wasn't suggesting that VHS does not have 480 lines, just that the amount of detail captured is not that high. That, it would seem, was completely wrong.

I think get it now -- Decreased quality from VHS (and any other source)only affects the horizontal because each line is stored on the tape linearly, right?

[trevlac]

What trevlac said. I think we agree 100% for once.

So you think Satstorm is crazy ??



@iantri,

Yes .. H quality goes down. V quality is always bad. This is a bit confusing ... most likely because of the poor words used like lines to mean horizontal resolution .... and so on.

BTW: This is all for a normal TV. Progressive and HD are different.

[junkmalle]

I think, in a sense, everybody is right about vertical resolution. Yes, all NTSC video sources contain 480 scanlines of information, all of which can be different. But in the real world video aquisition devices (cameras, film scanning equipment) are intentionally defocused (or blurred after scanning) so that they don't deliver that much resolution. This is to reduce aliasing, strobing, moire, and flickering distortions.

[lordsmurf]

So you think Satstorm is crazy ??

Okay, maybe 99%

video aquisition devices (cameras, film scanning equipment) are intentionally defocused (or blurred after scanning)

I would disagree. I wonder what BJ_M would say about this one.

[junkmalle]

Here is my simplified explanation of why you can't capture the full details of a ~352 line analog image with a digital frame of 352 samples (for example, capturing VHS at 352x480).

I'm not going to argue whether 352 is good enough or not, that's up to each individual to decide for themselves on a case by case basis. But I am going to show a simple (admittedly extreme case) example that demonstrates why sampling with 352 pixels will lose some small details. There are people here with all levels of familarity with the subject of analog to digital conversion so I will try to keep my explanations simple. I myself am not an expert on A/D conversion, but I think I have enough knowledge of image aquisition and processing to broach the subject.

Lets start by looking at the trail left by the TV's electron beam as it makes a single scan across the TV screen from left to right:



As you can see, this signal oscilates smoothly between black and white as it moves across the screen. It starts out a medium shade of gray, gets darker, then brighter, the darker again etc.

Let's represent this graphically as follows:



Superimposed on the electron trail is a graph of the intensity of the signal vs. time. In our example, we've scaled time so that it is equivalent to position. You can see the the low points of the graph correspond to the dark parts of the scanline, the high points on the graph correspond to the bright parts. I have arbitrarilly labeled the vertical axis of the graph from 0 to 100. 0 represents black, 100 white, and values in between represent the different shades of gray. This is the scale we will use for our A to D conversion.

Now, lets assume that the trail left by the electron beam was from a signal that oscilated between black and white 352 times as it traversed the screen. And that this is the analog resolution of VHS tape (this is a little higher than the 300 lines I measured on my VCR earlier but it's in the same ballpark and it works well for the rest of my example). The pattern is full black 176 times, full white 176 times and shades of gray in between.

Analog to digital conversion can take two basic forms. For slow signals you can take instantaneous readings and record them as numbers. For higher speed signals this is difficult to do so you end up taking a reading of the average of the signal over short periods of time. This is generally accomplished by counting the number of electrons that hit a sensor over that period of time. Lets say we design a sensor that can count the number of electrons that hit it in the time it takes our electron beam to scan 1/352 of the way across the screen. If no electrons hit the sensor during the sample time the signal is 0. If 50 electrons hit the sensor the reading 50, if 100 electrons hit the sensor the reading is 100. The value reported is equivalent to the average signal strenth over that sample period. We don't really know if we got a reading of 50 because 50 electrons hit the sensor all at once, or five groups of 10 electrons hit, or 50 individual electrons hit the sensor during our sample time.

Here's a graph of what our sensor detected over a short period of time:



As you can see, the previous graph has been broken up into horizontal sections. We're only looking at a small portion, but the entire graph covers all 176 peaks, and 176 troughs of the signal and has the total scanline split into 352 boxes. I've shaded the area under the wavey signal line gray because that represents the electrons we are collecting. The proportion of the entire box that is shaded is the value that is recorded. In the first box the signal starts out around 50, drops to 0, then rises again to 50. Overall, only about 25 percent of the box is filled in gray -- so the value assigned to this box (pixel) is 25. The second box starts out around 50 but quickly rises to 100 and falls back to 50. Well, it's kinda difficult to say exactly how much of the box if filled but it's about 75 percent -- our second pixel gets a value of 75. Since we are sampling the signal at the same rate that it is changing we simply get a digital result that oscilates between 25 and 75. There are 176 25's and 176 75's That's exactly what we wanted! We started with 352 "lines" in our analog signal and got 352 digital lines in our conversion. We've perfectly converted the analog signal, right?

Here's the "but...":

The analog signal we started with happened to have a nice alignment with our sampling boxes. But the an analog signal can go up and down wherever it wants. It may only be able to make the full swing from black to white 352 times as it scans across the screen, but it's not limited to when and where those transitions take place. Here's the same 352 line analog signal shifted to the side a little:



Over the first box it goes from 0 to 100. It covers... lets say... 50 percent of the box. In the second box it goes from 100 to 0 and it covers... 50 percent of the box. The third box is the same as the first, 50. The forth box is the same as the second, 50... Gee, all our digital samples now read 50. We have a featureless gray digital scanline -- not a nice one alternating between black and white! It was exactly the same analog signal, just shifted slightly to the side. We got two completely different results. Where did the lines go?

This is of course all "theory". And I've picked a case which I knew would be very bad. In the real world you won't see patterns exactly like this very often. But you will see some manifistations of this with more common images. Imagine a small branch of a tree against a white cloud in the sky. It could end up like one dip in the samples above. In one frame it might line up so that it covers a single pixel and shows up as one very dark sample in the middle of a bunch of light ones. But as the camera pans it might end up being split between two samples and end up as two slightly dark pixels against a white background -- it has blurred. If the camera is panning slowly it will strobe between one very dark pixel and two slightly dark pixels. How about the pinstripes on someones suit? The stripes on a soccer goaly's shirt. Any sharp transition between light and dark...

[junkmalle]

So you think Satstorm is crazy ??

Okay, maybe 99%

video aquisition devices (cameras, film scanning equipment) are intentionally defocused (or blurred after scanning)

I would disagree. I wonder what BJ_M would say about this one.

I'm not saying the cameraman intentionally defocuses the camera, but that the camera is designed not to be able to focus up the the electronics full ability to resolove.

[trevlac]

Yes, all NTSC video sources contain 480 scanlines of information, all of which can be different.

I don't follow you here. Let's break it down a bit.

Interlaced is actually ~240 scan lines 60 times a second. This 240 basically equates to the vertical sampling rate. This limits the vertical resolution of all NTSC source.

If a DV cam or scanner or whatever aquires more vertical detail ... it must be filtered (blurred / distorted / whatever) to fit into 240 scan lines.

So ... the aquisition process may limit the resolution ... and ... the standard already limits the resolution ... but ... any format for the given standard (VHS/SVHS/DVD) has the same limits.

In the End .... if you are saying different methods of aquireing an image produce different results ... probably. But that result is the same for DVD/VHS/DV ...

PS: What the heck do I know. But that VHS resolution pic is damb compelling info.

Oh Yeh ... and another thing ... pixels are not little squares.

[junkmalle]

Yes, all NTSC video sources contain 480 scanlines of information, all of which can be different.

I don't follow you here. Let's break it down a bit.

Interlaced is actually ~240 scan lines 60 times a second. This 240 basically equates to the vertical sampling rate. This limits the vertical resolution of all NTSC source.

Yes there are 240 scan lines per field. But two fields are interleaved together to make 480 spacially separate scan lines over the hight of the display. On a computer I can generate a image of, say, alternating black and white, single pixel thick, horizontal lines. Then I can burn that image as an MPEG file on a DVD. When you watch this on TV it will flicker like hell but you will be able to see every scan line. If you record it to a VHS tape and play it back you will still see all 480 lines. But in the real world you won't find any image taken with a video camera (analog or digital) that can clearly show the 480 alternating black and white lines.

Oh Yeh ... and another thing ... pixels are not little squares.

I'd put it this way: "pixels have neither shape nor size. They aquire those traits when the are displayed or printed." Of course, the sensors on a video CCD have size and shape. It's not exactly like you would expect though. But that's another topic and I don't want to spend another half a day writing it up! Here's a site with some good descriptions of imaging sensors:

http://www.tasi.ac.uk/advice/creating/camera.html

[trevlac]

@junkmalle

I do like your example and pictures ... but I disagree with a few of your points.

1st ... On the V res thing. I'm not sure if we are disagree at all. Might just be talking about different points. I'm saying that for NTSC ... the V res is always the same. VHS, DVD, DV, SVHS. Lots of ways to give that number, but it is the same.

2nd ... On the H res thing and sampling ... the points I disagree on are:

a) You are showing that the apature is as wide as the time between samples. I don't think this is ever the case. And for sampling an electrical signal (aka video capture) I really don't think this is the case. I'd say it is even reasonable to say there is zero apature effect.

b) Samples(pixels) do not have to line up with peaks.

The theory
3MHZ is the max bandwidth of VHS. 240 is a measure of that bandwidth in H TVL. 320 is TVL adjusted for a 4:3 screen. If you can see 300 TVL from your VHS, you see less than 320. 320 'Fits' into 352 because it is less than 352 and they are effectively the same thing. Translating 320 to bandwidth you really need to know the length of the sample line. Assuming 53.333msec (the digital standard) 320/53.3333 = ~ 6MHz. This is 2x 3MHZ. The sampling theory states you need to sample at a rate > 2x the highest frequency to be able to recreate the analog. 352 is > 2x the highest frequency in VHS + some headroom.

The reason you can recreate the analog with so few samples and why samples do not have to line up with peaks is that some complex math and what is called the sinc function is used to calculate the points on the wave between the known sample points. sinc is sine(pi*x)/pi*x IIRC. A Lanczos resize is a practical implementation of a sinc function. When you resize, it figures out the new sample points. And it's pretty good.

The Practice
Capture cards all sample the electrical signal way above the horizontal numbers we are talking about. Try 1500+. They then down size in digital (mathematically).

You need some headroom for this to really work. But 352 should be sufficient headroom above the 320 number which is probably never reached in any VHS tape.

The reason for the headroom is that the cards need to resize fast, so they can't be messing with the perfect resample. Their algorithms will remove some of the detail. The more headroom the less they will effect the detail.

You could cap high and resize latter or cap low. You might get the same results. Depends upon the resize methods.


A Test
A test on specific hardware/ methods would be to record the rez chart to VHS. Capture at 704 and at 352. Lanczos resize the 352 to 704. See any difference in the wedges that measure H res? If no ... zero problem. If yes ... you then need to consider if your source will ever have 3MHZ.

[trevlac]

I'm sorry but I can't resist talking about this one.

Oh Yeh ... and another thing ... pixels are not little squares.

I'd put it this way: "pixels have neither shape nor size. They aquire those traits when the are displayed or printed." Of course, the sensors on a video CCD have size and shape.

It is fair to be clear on what I mean by pixel ... because I guess you could think of the elements in a monitor as pixels ... or the sensors in a ccd maybe. But I mean the pixels in a digital image. The samples.

So I like your short definition. But I must say more.

The trap people fall into is thinking of pixels are little squares or rectangles. This causes too much confusion. If you take a 3cmx4cm picture and measure it at little points ... you effectively have samples/pixels. If you measure again but this time closer together you again have samples of the same picture. These measures don't have a size. They do have an amount of space between them. So ... 352x480 and 480x480 and 640x480 and 720x480 can be basically the same picture with different measurements. The 720x480 one can know more detail because the measurements are closer together. A black picture can be measured in 2x2 as effectively as 720x480.

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

On the CCD thing. I havn't really read much on those ... I will read your link. Thank you.

[junkmalle]

On the H res thing and sampling ... the points I disagree on are:

a) You are showing that the apature is as wide as the time between samples. I don't think this is ever the case. And for sampling an electrical signal (aka video capture) I really don't think this is the case. I'd say it is even reasonable to say there is zero apature effect.

It doesn't matter. Take the middle 50 percent of the aperture, even take a point source from the center of each aperture or any other fixed interval. You will see the same distortions (described below).

b) Samples(pixels) do not have to line up with peaks.

Of course not, that was my whole point. I chose those two algnments as extreme examples. Repeat it with any any phase and aperture size you like. You will see the following. As you slide the wave sideways the picture oscillates between high contrast black and white pixels (my first example), through less and less contrast between the pixels, and finally at 90 degrees (1/4 wave) the contrast is completely gone (ie, the solid gray in my second example). It then starts to cycle back to more and more contrast between the pixels until you reach the next peak in contrast at 180 degrees. I leave it as an exercise for the reader to see what happens when the signal frequency differs by a small amount!

3MHZ is the max bandwidth of VHS. 240 is a measure of that bandwidth in H TVL. 320 is TVL adjusted for a 4:3 screen. If you can see 300 TVL from your VHS, you see less than 320. 320 'Fits' into 352 because it is less than 352 and they are effectively the same thing. Translating 320 to bandwidth you really need to know the length of the sample line. Assuming 53.333msec (the digital standard) 320/53.3333 = ~ 6MHz. This is 2x 3MHZ. The sampling theory states you need to sample at a rate > 2x the highest frequency to be able to recreate the analog. 352 is > 2x the highest frequency in VHS + some headroom.

Yes, this is the Nyquist theorem. But that doesn't mean that at 1.99x you have crap and and 2.01x you have a perfect representation of the original analog waveform. It is a rule of thumb that says the highest frequency that can appear is 1/2 the sample rate. It doesn't say that it takes a 2x sample rate to perfectly represent the waveform.

Capture cards all sample the electrical signal way above the horizontal numbers we are talking about. Try 1500+. They then down size in digital (mathematically).

I was unaware of this detail. But that is more or less what is described in my collection aperture examples. (Minus any resizing filtering that goes on within the hardware.)

A test on specific hardware/ methods would be to record the rez chart to VHS. Capture at 704 and at 352...

I was going to do this but didn't have time today. I may tomorrow. The problem though is that I can only test one particular capture card. And as we all know from earlier discussions that cards handle things differently. Anything I find may be limited to my card -- and since it captures directly to MPEG (Hauppauge WinTV PVR 250) we have that additional complication.

In any case Trevlac, thanks for the perfectly civil discussion. I've seen so many threads like this that break down into huge flame wars!

[BrainStorm69]

EDIT: Note that these caps are straight from DVD -- I will try to post the same caps from a VHS recording of the same tonight

I thought we had been all through this some time ago in various other threads, but here we go again.

OK, someone asked for pics. I'm going to post two screen caps from each of 3 cards...my ATI7500 AIW (Rage Theatre), my Avermedia AverTV Stereo (BT878), and my Avermedia EZDVDMaker PCI (Philips SAA7130). I captured from each at 352x480 and 704x480 in AVI format using the huffyuv codec. I used MMC 8.9 for ATI and VirtualVCR for the AverTV Stereo and EZDVDMaker. Then I used TMPGEnc to render each to a 352x480 mpeg2 with the exact same settings. I screen captured with Virtualdub mod in tga format and I then used PaintShop Pro to convert to jpeg using the exact same compression setting (9).

I'll let everyone make their own assessments...

ATI 352x480


ATI 704x480


Conexant/BT878 352x480


Conexant/BT878 704x480


Philips SAA7130 352x480


Philips SAA7130 704x480

[Wilbert]

Great test!

I guess I have to apologies to LordSmurf a bit. There's a difference in the ati screenshots, but it's not as bad as BT878 (at 352x480).

Could you make two more caps at 368x480 and 400x480? The differences between the screenshots should be different with these sizes.

[trevlac]

@BrainStorm69

Nice post. Personally I think you should show 352 resized up to 704 via lanczos vs a 704 cap. The TMPGEnc resize is probably a bicubic which may both sharpen and blur.

But this goes to my point about practical. Test your full process and see what works for you. If you are going to make 352x480 dvds, looks like a 352 cap is better.

BTW: The BT878 352 looks so bad because there is a vertical filter. Try 360 vs 368 on the BT. BIG difference not due to 8 pixels.

Also ... it is reasonable to cap at 368 with the BT and BTWincap drivers ... and just crop to 352. The AR will not be off as much as you would think. This is because the BTwincap (NTSC) caps about a 0.6 wider window than is used in the 352 size. This translates to about 4 pixels. So the AR is off by about 12 pixels not 16. VirtualVCR lets you crop as you cap.