How and what values to use when resizing for OVERSCAN

[FulciLives]

Hello

Recently been playing around with doing some TV captures using an analog type TV tuner capture card (the AverTV Stereo).

I capture with iuVCR and encode with the HuffyUV codec. I capture at 352x480 (my computer is just a touch slow for full D1 resolution) and I am attempting to use TMPGEnc to convert to DVD using 352x480

I'd like to resize the capture JUST A BIT to off-set overscan.

I tried doing this with an AviSynth script but it seems to screw up my field order NO MATTER which field I pick in TMPGEnc.

I've tried the custom size VIDEO ARRANGE METHOD in TMPGEnc but I'm not sure what would the correct values should be.

I've been using 336 for the width but I'm not sure what to use for the height in order to retain the proper aspect ratio.

My calculations say to use 336x458 but that seems to put the image up slightly higher (more black below the image than above it) so I don't think actually works. I've tried 456 and that seems to be better but then the aspect raito is off although it appears to be a very small amount (I sure can't tell otherthan by calculating the numbers).

So does anyone know what figures I should be using to resize my 352x480 capture for overscan while encoding it to MPEG-2 at 352x480?

Thank you!

- John "FulciLives" Coleman

P.S.
I don't want to "overdo" the overscan off-set. I just want to help it a bit. I'd hate to actually overdo it and risk seeing some of the black border on the TV.

[Suds]

I am assuming you are using AVS 2.5x.

http://www.nic.fi/~lhahne/GripFit_YV12.zip

[DJRumpy]

I use 16 pixels per side, for a total of 32 pixels.

480 - 32 = 448

Just make sure the letterboxing added is no greater than 16 on both sides. As long as you don't crop by odd values, then your field order should not change.

[FulciLives]

I use 16 pixels per side, for a total of 32 pixels.

480 - 32 = 448

Just make sure the letterboxing added is no greater than 16 on both sides. As long as you don't crop by odd values, then your field order should not change.

So you are saying (if I understand you correctly) that I should use 16 on all 4 sides thus making my 352x480 capture into 320x448?

What if I capture at full NTSC D1 resolution of 720x480?
Should I just reduce by 16 all around as well?

What about doing 8 all around? Will that work as well?

I do want to minimize the overscan but I don't want to "risk" actually seeing any black either.

Plus I want to retain the original aspect ratio (or as close as I can get).

I tried using GripCrop or whatever via MovieStacker (a utility that comes with the KVCD stuff that creates an Avisynth AVS script for you) but it always seemed to screw up the field order so I've been using the CUSTOM SIZE in TMPGEnc for the VIDEO ARRANGE METHOD which let's me pick any size I want then centers it with black all around. This seems to work without screwing up the field order.

- John "FulciLives" Coleman

[vitualis]

Why do you want to adjust for overscan on broadcast material?

All broadcast material has overscan in mind when it was created so there is usually a boundary of about 10% where important stuff doesn't happen. I suggest that you just leave this alone.

Regards.

[FulciLives]

Why do you want to adjust for overscan on broadcast material?

All broadcast material has overscan in mind when it was created so there is usually a boundary of about 10% where important stuff doesn't happen. I suggest that you just leave this alone.

Regards.

Maybe it's just ME but I swear that if I capture at 352x480 and then encode to MPEG-2 at 352x480 and burn a DVD then watch that DVD on my TV ... well ... the image is ever slightly more overscanned than the direct cable feed. This is easy to tell because of the logos that many stations (such as SciFi which I record from alot) put in the lower corner of the screen. Switching back and forth from my DVD to the cable feed I see a (very slight) difference with the DVD being slightly more overscanned than the direct cable feed.

- John "FulciLives" Coleman

*** EDIT ***
Edited due to my dumb ass making stupid spelling mistakes hehehe

[ronnylov]

Have you tried FitCD?

On my TV I get too much shrinking with 16 pixels on left and right when using 352 pixels width. I can see some black on one of the sides. It's better with only 8 pixels horisontal. 16 pixels is OK on the vertical resize (when I'm using 352x576 PAL) but you should also keep the aspect ratio. That's why I use FitCD, becaue it's a great tool to calculate overscan rezising. It can generate the Avisynth script needed.

[DJRumpy]

So you are saying (if I understand you correctly) that I should use 16 on all 4 sides thus making my 352x480 capture into 320x448?

What if I capture at full NTSC D1 resolution of 720x480?
Should I just reduce by 16 all around as well?

Yes, you would reduce a bit on all sides to minimize overscan. You end up with a display, much like your PC monitor (most people leave a slight black border). The value to use as a letterbox varies a bit from TV to TV. You can certainly do 8 if that's what you want. I would suggest a few small test clips with RW media if your not sure. Once you get the right amount, then it will stay consistant regardless of what foramt your displaying (it's a display specific thing).

vitualis, I agree with you here, but there is one situation where I do resize to leave in a border, and remove overscan. It's specific only to broadcast captures. I'll use Sci-Fi channel as an example, as that is where the majority of my captures happen. They often broadcast in widescreen letterbox (4:3 aspect). When restoring it back to 16:9, the SciFi logo will be chopped off a bit if you cut right at the image/letterbox border. I find that I have to reduce the image by 16 on top and bottom to get everything to fit and display properly, while not showing any borders. Granted, it's a cosmetic thing, and usually unnoticable in under non broadcast capture situations. We're talking only 32 total pixels per side afterall.

For DVD conversions and such, I don't bother.

[LisaB]

I don't see how cropping by the same on all 4 sides could possibly maintain aspect ratio, except in the case of NTSC SVCD resolution. NTSC SVCD resolution is 480x480, and thus a square, so if you resize to 448x448 and add 16 pixel borders all around, then you can rest assured that the original aspect will be retained. But clearly this method doesn't work for any other resolution.

For 352 by 480, if you crop 16 off the top and bottom, then I think you should crop 352/480*16=12 off the sides (well it's 11.73.. but twelve will be the closest integer). So, I think resize to 328x448, then add borders 16 on top and bottom, 12 on sides. It actually doesn't matter what resolution the source is, as long as the *aspect ratio* of the source is 4:3. If you have a VGA aspect ratio source (like an AVI...they also call this 1:1 aspect, but I find that very misleading, unless of course you're dealing with a square AVI!) or a 16:9 source then things get much more complicated, and I don't know how to calculate...use FitCD I guess.

[energy80s]

If you've any sense you won't mess about with cropping for overscan as sizes vary from one TV to another, and are non-existant on PC monitors and projectors. Just keep the whole damn lot and be done with it! Oh and all broadcast material is already masked at 702 for transmission anyway, so you really aren't loosing much extra in the overscan area.

[DJRumpy]

Although I'm sure FulciLives knows that overscan reduction isn't necessary, he (or she... sorry FulciLives, I haven't looked at your profile ) specifically asked how it's done. Telling them not to do it doesn't answer the specific question.

LisaB, to be specific, we're not cropping. We're reducing the entire image area, and filling that 'free' area with letterboxing. You should also keep in mind that an SVCD (and CVD) is stretched out to a 4:3 aspect ratio on playback. It does not remain 480x480 or 352x480. The full horizontal width on playback is around 640 pixels ( 480x1.33= 640 rounded ), meaning your math is flawed. I would gamble that the majority here wouldn't be able to spot a 4 pixel difference, or even a 30 pixel difference from the vertical and horizontal though. Your making it harder than it needs to be. You can change the aspect ratio quite a bit before any appreciable change in playback aspect would be noticable.

FulciLives, I would suggest you experiment with different values. You won't notice such a small change having any affect on your display ratio. Start with a small value, and increase it until your happy with the result. Don't worry about a few pixels difference from top to side. If you want to be exact (to each his own...), simply use the aspect ratio of your TV as a guide. If you crop 16 pixels from the height, and your TV is a 4:3 television (1.333 decimal), then multiply 16x1.333 to get the amount to letterbox on the horizontal ( 16x1.333=22 rounded by 2 ). Remember a 4:3 television is 1.333 times wider than it is tall

[energy80s]

I still don't see the point, even though I know that Kwag does a lot of this resizing/cropping for his KVCD formats. Keep the horizontal at either 352, 704 or 720 and you'll not go wrong.

[DJRumpy]

FulciLives does it because he/she wants to. Putting the maximum amount of video on your display, without borders. A simple reason, and certainly not everyone's bag of tea, but a valid one.

To each his own...

[LisaB]

DJRumpy,

You're right, I was using the wrong terminology...crop isn't correct, I should have said "shrink 16 off the top and bottom" instead of "crop 16 off the top and bottom"! But I do get it right in the next sentence when I say "...resize...then add borders"

But, I beg to differ regarding the math.

For 352x480, if you shrink 16 off top and bottom and 12 off sides, then the image area will be 328x448. As you say, the horizontal is stretched from 352 to 640 for display purposes, or in other words the horizontal is scaled by 640/352 while the vertical is left alone. Thus, the final image area for display (i.e., on the 4:3 TV) will be:

horizontal = 640/352*328 = 596.36
vertical = 448

So, display aspect of the image area will be 596.36/448 = 1.33 !!

Thus, shrinking 16 vertical and 12 horizontal really does preserve the aspect ratio.

The moral of this is that as long as you resize so that the new frame size has the same real aspect as the original frame size (i.e., 328/448 = 352/480), then your display aspect ratio will be preserved. That is why a square frame (i.e., 480x480) should have equal amounts shrunk off each side, while non-square frame sizes shouldn't!

[adam]

nevermind.

[FulciLives]

Hello

First let me say that I am very happy and glad with all of the responses here. In other words I appreciate the help.

I do hate to quote myself but several people have asked why I want to do this with a TV capture. Apparently some people don't read very well ... here is a quote by me that appeared early in this thread:

Maybe it's just ME but I swear that if I capture at 352x480 and then encode to MPEG-2 at 352x480 and burn a DVD then watch that DVD on my TV ... well ... the image is ever slightly more overscanned than the direct cable feed. This is easy to tell because of the logos that many stations (such as SciFi which I record from alot) put in the lower corner of the screen. Switching back and forth from my DVD to the cable feed I see a (very slight) difference with the DVD being slightly more overscanned than the direct cable feed.

In other words if I capture at 352x480 (I use AVI capture format) and then convert to MPEG-2 at 352x480 the image does not 100% match the cable feed. For whatever reason it is ever slightly more overscanned. I can see this easily by switching back and forth from the cable feed to my DVD using the exact same TV. Of course the program content is different but the channel logo is in the same place. The only time Sci-Fi channel (my frame of reference since I capture from that channel alot) seems to change the location of the logo is on widescreen presentations. It always seems to be in the same place with full screen programs but I've noticed with widescreen content they move it up so that at least part of the logo is in the actual image. I suppose they do this so if you try to crop the widescreen image you will still see (at least partly) the logo (unless you actually cut a few lines INTO the image).

Anyways I think I now agree that doing the same pixel size around all 4 sides will probably not throw off the aspect ratio very much ... at least it will be so small (in error) that you really will NOT be able to see it. However I am interested in the math that LisaB came up with for 352x480 but what about 720x480 captures LisaB?

Thanks again all and hope I hear something from you LisaB about 720x480 math for resizing.

Also remember I don't want to COMPLETELY eliminate the overscan. I don't WANT to risk seeing any of the black. I just want to off-set the overscan a bit so I see at least a bit more info.

And DJRumpy ... I'm a male sorry if that is a disappointment hehehe :P

- John "FulciLives" Coleman

[LisaB]

I should preface this by saying that I'm rather interested in this subject, and haven't really found any clear explanations of overscan calculations out there...so I'm just going by what I know, and using reason. I rather hoped Adam had said something (he seemed like he wanted to)...since he's given very good advice in the past. Also, I would like to know where DJRumpy got his info, as he has given extremely good advice about IVTC in the past..

Anyway, DJRumpy threw out the number 640...but the logic of resizing doesn't depend on this number. Consider again the case 352 by 480.

1. the 352x480 frame is distorted so that it displays on your TV with a 4:3 aspect ratio
2. It doesn't really matter if that is done by stretching horizontally (as DJRumpy says is the case), shrinking vertically, or even some combination of the two.
3. Since the 352x480 frame undergoes this distortion, any area *within* the frame will undergo the exact same distortion.
4. In the case of a frame adjusted for overscan, it will have an image area surrounded by black borders.
5. The image area is just an area within the frame. This means it will undergo the same distortion as the frame.
6. Thus, as long as the width/height ratio of the image area is the same as the width/height ratio of the frame, then the image area will *also* display on your TV with a 4:3 aspect ratio. This is what we want, and this is what it means to preserve aspect ratio.
7. so, if you shrink by 16 on top and bottom, the new height is 480-16*2=448.
8. so, we want the new width such that: 352/480=width/448
9. so, width =448*352/480=328, which is 24 less than 352.
10. so, you want to resize to 328x448, then add 16 pixel borders top and bottom, and 12 pixel borders left and right.

For 720x480, if you shrink 16 pixels on top and bottom, then width=448*720/480=672; and (720-672)/2=24; thus shrink 24 pixels left and right. Those are the borders to add. You of course want to resize to 672x448.

[alenhard]

Again, the way an image/movie should be resized is entirely dependent upon the TV you will be watching it on.

I have a Sony that displays everything perfectly except for about 20 pixels on the right-hand side of the screen only. I can add a bar to the side of an image on a slideshow that I can't see on my TV, but that is visible on my parent's TV (while theirs seems to take a little off the top and left-hand sides). Weird, I know.

You'll probably have to mess around until you find what works for you.

[LisaB]

Oh, and by the way FulciLives,

Your calculation of 336x458 is good, and will preserve aspect ratio. The problem with this is that to get it truly centered would require 11 pixel borders top and bottom! TMPGEnc is not going to do that, but will instead use a 10 pixel border and a 12 pixel border--I think there is some requirement that the added borders be even number of pixels.(?) This is, I bet, why it looks off...better to use more round numbers.

[DJRumpy]

You do not crop or add vertical values by odd numbers, so you can prevent the field order from changing. You won't notice a single pixel width offset.

Good point regarding the distortion. Any changes you make to the horizontal, or vertical, would indeed be reflected when the image was stretched out to it's full width on playback. For SVCD, it would be logical to assume that you would cut even amounts from each side, which would then be magnified when the image is stretched out on playback. CVD would require you to multiply your horizontal value by .733 to keep the aspect ratio after the image had been resized. That smaller value would be magnified to a number that was greater than the value used for the vertical on playback. Still just busy work to my mind. You won't see a difference in aspect ratio for a 4 pixel difference

[lordsmurf]

TMPGEnc has the video arrange method. Use "full screen - maintain aspect" and cut a few off each side. It'll maintain aspect, and the cutting away of pixels will slightly stretch the image.

I wouldn't do more than 10 per side (it'll actually be about 12 on top, 8 on each side when maintaining aspect).

The root of the problem is probably your Aver card. I've seen that too, and only on that card.

[FulciLives]

Just an update here ...

I've decided to capture at 720x480 but some stuff I think is better off at half D1 (352x480) so when I decide to do half D1 I use the CENTER (custom size) for the VIDEO ARRNAGE METHOD in TMPGEnc then I am resizing to 340x464

Following LisaB's math:

352/480 = X/464

464*352/480 = 340.266666

Not 100% accurate but pretty close.

Also consider this:

352/480 = 0.733333
340/464 = 0.732759

That's pretty darn close.

I decided to go with 340x464 instead of 328x448 because although I do want to compensate for overscan I really do NOT want to overdo it. I'd hate to overdo it and RISK seeing some of the black border (depending on the TV).

Thanks again all

- John "FulciLives" Coleman