704 vs 720, why crop?

[dphirschler]

fixed. -- Darryl

[Alan69]

What is more important to you? Correct aspect ratio or 16 more horizontal pixels in your DVD Standalone playback on TV? Faster encodes/smaller filesizes or the opposite?

But remember to factor that shift in algorithms back into this part of the statement. It isn't just 16 more pixels for a ~2% change, it's 16 more that causes a major shift in how the pixels are calculated. 2:1 stays within the original pixel boundaries, and 3:1 samples across the '2 pixel' boundaries of the original picture with sliding percentages. The 2% of extra pixels shifts everything else, and could cause a more than 2% total change in the picture after carrying through everything else.

With 352 being right at the border anyway, even a few percent difference in picture quality might be noticed. If it actually ends up a 4 or 5 percent shift, it would be a very good thing to pay attention to. I have more than a technical interest, I consider 352 to be just around 'not passable'. Even if the difference itself is there but not quite good enough on it's own, it may move things enough to say hell with it and drop the standard a bit. As it is now I consider 352 around a 60% grade with 70% passing, just a little too poor to even stretch the standard and use it anyway. If I can get it to 65% or so I'd likely say screw it and use it a lot more anyway for general things.

Various differences could trade off, and produce little real difference between the two for PQ. Or the trade offs could reinforce each other, and combine to make a noticable shift in sharpness vs smoothness between the two methods. Basically if there's a difference, then both the resize and the sharpen/blur difference get to happen in the ~700>352 transform at the same time, which is generally better than sequential operations since the errors don't compound. Checking it out and knowing which way does what could make 352 more useful.

Time to do some close testing from a good analog source.. 720 to 352 both ways and a direct 352 cap all of the same cartoon should make it easy to see.

Alan

[trevlac]

Alan,

How are you going to resize? VDub,Avisynth ... Lanczos, Bilinear ..

It really makes a difference. As I recall, the Theater chip ATI's do a bit of bluring with a downsize. Looks like Bilinear.

The doom9 capture guide has a bunch of example pictures from about 4 different chips. Look in section #5. BTW: The BT878 cards kick the ATI's but on a resize, but it screws up the vertical if you go below 368 Horizontal. A bit of a hollow victory.

Edit

Here is a good resolution test pic. 720x480 this time. Don't resize it for a 704 test. Crop. The 6.75MHz circle is the max for DVD. If you can see lines in it, you have good detail. A 352 frame can hold a max of ~ 3.3MHz. So all circles but the 3MHz one should be grey. If they are not, you have aliasing (where the details have changed and are not being represented "properly"). The ATI will filter out these high frequencies. The BT for example may not. I don't think a lanczos or Bicubic downsize filter. Bilinear does.

As a note. The 'wedges' are read by finding the number where you can no longer reasonably see the lines. This is the measure of TVL resolution. For NTSC, 80 TVL =~ 1MHz. The up and down one measures Horizontal res. The side to side one measures Vertical. Vertical should be 480. Even for VHS. If you go less than 368 on the BT you loose vertical res. This is bad.



Here is an example of a 368 cap on a BT. Some may say it is sharp. The detail in all but the 3MHz circles are artifacts. Not that they are bad. I like this better than a soft picture. Unfortunately there is an artifact in the test image wedges. So it is hard to tell the amount of xtra artifacts you may be getting if a downsizer does not filter. The wedges should be used at full res.

[LNielsen76]

So I did my last test:

Crop to 704x464
Mask to 756x512
Resize to 352x480
Encode at 352x480

This wasn't worth it. Results came out too blurry. To much quality sacrificed. I was able to display a greater number of horizontal pixels in Standalone Playback, but the image quality blurred way too much.

@Alan

I can see now that where aspect ratio is preserved by crop and then resizing, image quality is sharper, encodes faster, and file sizes smaller. This may be worth it even though 16 horizontal pixels are sacrificed in the DAR in Standalone playback.

However, cropping to 704, masking back to 720 and resizing to 352 allows for faster encoding and smaller filesizes as compared to a direct resize to 352 from 720. Also, 16 extra horizontal pixels are displayed in Standalone playback. But, the image is blurrier than maintaining PAR.

It is suggested that the 352 resolution is below the min 368 for acceptable playback quality and it is probably best to simply use full D1. However, eventhough cropping produces sharper results than resizing, more image is displayed by resizing. I like to think the additional (minute) blurring as additional noise reduction (though not proper NR.)


Summary: Converting 720x480 to 352x480

Resizing: Crop to 704, mask to 720, resize to 352

16 More Horizontal Pixels in the DAR, but blurrier results.

Cropping: Crop to 704, resize to 352

16 Less Horizontal Pixels in the DAR, but sharper results.

[petar]

I'd like to summarize what's been said and see if we can get to the bottom of this.

If I understood trevlac and wilbert correctly, as far as hardware DVD players are concerned there's only one pixel size for 704 and 720 and the true 4:3/16:9 is for 704 (more-less, since it's 702 for PAL and 711 for NTSC).

This would mean that there will be a difference in the aspect ratio between software players (which resize 704 and 720 to the same width on playback) and hardware players (which don't display 704 and 720 in the same width because they use the same pixels for both). OK, I know the difference is not big, but it confuses me that there's a difference.

And is this true for all hardware players? I'm not an expert in any way on electronics but what about the newer players that can play DivX and SVCD? I presume they'll be capable of resizing on playback, similar to the software players, in order to accomodate all these formats. Also X-Box and PS2 - they can also resize, what about them? Do they all also display 720 as chopped 704 or do they act as the software players?

[lordsmurf]

As soon as you consider all the players that properly parse 544x480, 640x480, 320x240, etc etc .... you do realize all these theories in this thread are completely full of holes and out the window?

Too many of you are confusing and mixing otherwise separate concepts.

Oh well, have fun.

[Alan69]

As soon as you consider all the players that properly parse 544x480, 640x480, 320x240, etc etc .... you do realize all these theories in this thread are completely full of holes and out the window?

Too many of you are confusing and mixing otherwise separate concepts.

Oh well, have fun.

Since when does how many people are confused by something have any bearing on whether something is valid or not? Many people are confused by many advanced topics, that doesn't mean that advanced topics don't exist.

If you choose to believe it's nonsense, then that is your choice, knock yourself out. But in that you are incorrect.

Other resolutions have nothing to do with it. 352 pixels edge align into 2 704 pixels, and do not edge align with 720 pixels. This will cause a change in the sharpness of the picture, the extra pixels are put into the 352 picture by blurring them in. And it's going to show up on any player when you play back that 352 encode since it was permenantly put into the output by which way the encode was done.

[Alan69]

It is suggested that the 352 resolution is below the min 368 for acceptable playback quality and it is probably best to simply use full D1. However, eventhough cropping produces sharper results than resizing, more image is displayed by resizing. I like to think the additional (minute) blurring as additional noise reduction (though not proper NR.)

Yep I'm thinking the 720 resize to 352 may be preferred for most cases. Cartoons and similar where edge detail is important may be better with the crop to 704 first though.

I've got a ton of aviation footage in 720 that would be really nice to put more per DVD at 352 if I can make it at least 'OK'. Time for some thorough testing, but I have the gutters off the house repairing and repaintinig the eaves, so it will be a day or two before I can mess with it much. Real life sucks sometimes. At least the house is nearly repainted in Liquid Ceramic and sealed now, a few more weeks and it won't need painting or repair again for 30 years.

[lordsmurf]

352 pixels edge align into 2 704 pixels, and do not edge align with 720 pixels. This will cause a change in the sharpness of the picture, the extra pixels are put into the 352 picture by blurring them in. And it's going to show up on any player when you play back that 352 encode since it was permenantly put into the output by which way the encode was done.

That depends entirely on the resize method.

[vhelp]

Here it is again, my recommendations:

If your "captured" source resoltuion is:

* 704 pixels, and your goal is 720 DVD compliance (dispite the 704) then
... one should resize by PADDing 8 pixels (left/right) to obtain 720 resolution.

* 704 pixels, and your goal is to bring down to yet another DVD compliance, by
... resizing down to 352, then just resize to 352 pixels.

* 720 pixels, and your goal is to bring down to yet another DVD compliance, by
... resizing down to 352, then this is the best method, and maintains the Aspect
... Ratio. We crop off 8 pixels from the Left and Right sides to make 704, then
... we resize down to 352 resolution.
... .
... (A special note of consideration here.. If you crop** off 8 pixels on each side
... of the source, these 8 pixels are pixles you would never see on your tv set.
... if you dont' crop off these 8 pixels from your 720 captured source, and you
... proceed to resize down to 352 (720 downto 352) you will in effect, blur your
... source, and possible affect your Aspect Ratio as well. The problem with
... this approach w/ many people here, is that they do not properly resize and
... encode w/ proper "Orientation" technique. TMPG for instance, has a wonderful
... Video Arrage Method feature that is over-looked. It would be hard
... for me to explain here HowTo properly Setup and Use this area of feature, so
... I won't go into it here But this is obviously an area where most people
... get messed up in.., because they don't have a handle on Arranging their video
... source when they encode. This is no wonder why we all get confused in a few
... things. CCE does not have such a feature, but you can alternatively use a
... AVIsynth script to accomplish the same thing as TMPG's Video Arrage Method)
... .
... On another note of resizing. There are many AVIsynth filters for Resizing.
... Many will blur or sharpen, or use various sub-sampling / inter-polating
... techniques while resizing. These all effect the final look of the
... AVI source.

* 688 pixels, and your goal is 704 (because going to 720 would only blur your
... final encoded MPEG) then PADD by 8 pixels (left/right) to obtain
... the 704 resolution.

* 688 pixels, well.., you can PADD to 720, but that might give you an un-desireable
... looking MPEG when played on your pc. Remember, our goal is to closely match
... what the pros do when they produce commercial DVD's. In my OP, "masking" 16
... pixels to the Left and Right sides of the final encoded MPEG source will look
... undesireable to me. I my source MPEG to look as close to those done commercially.
... 8 pixels (left/right) is enough. Any more, and you are over-doing it, making
... your source look ugly or strange.

* 640 pixels is a tough one to handle here. However, I would resize down to 480 or 352.

crop** - Also, I wanted to explain briefly here, that there are yet other
techniques to resizing your source to a final conclusion. And that is from inside
a "Capture" process. For example, if your source is Widescreen, you can take advantage
of your capture app's cropping (assuming your capture app has one)
As of this writing, I am now becomming accustom to using iuVCR w/ my Winfast card.
I like it because of it's Segmentation abilities (under Win98 Gold's 4gig limit)


Winfast TV2000 XP "Expert" capture card - REV C
(another revision was made to this board)

I also have a slight doubt about the various Resolutions to use (which one is best
to use in a given capture project)
.
I've ben using 688 for a while now, and I'm not sure if that's "really" the proper
resolution to use for this card. All I do know is that when I capture at various
resolutions, they all look great when I'm done finalizing them to an MPEG. Which
leaves me wondering about those peoples out there, weather or not they are
processing/encoding their sources effectively, leaving doubts in their minds on
weather or not to use a different resolution.
.
There ought to be a rule for each capture card, in their proper resolution setup.
I'll have to continue my tests, which I'm far behind in. I want to eliminate using
AVIsynth in my encoding process, when my source is true film in nature, and just use
virtualdub. My prefered technique is to Post Process my source. But, my problem
with this is that even though I use the Huffy codec in my initial capture project,
when I Post Process it to another .AVI (my IVTC process) to a Huffy, it's converting
again, the source. I don't want to re-convert my source. Saving to Huffy again
is re-converting the source. What I want is NO RE-CONVERTING what-so-ever. And as
I have said elsewhere's in another thread, there is no such a codec in existance.
Anyways.


So, to summarize:

My recommend would be like this:

If your capture source is 688 pixels, padd to 704.
If your capture source is 720 pixels, then leave it that way.
If your capture source is 704 pixels, then you can do either of THREE things.
-- 1) padd to 720, or
-- 2) leave it as 704, or
-- 3) you can resize down to 352

Cheers,
-vhelp

[Alan69]

352 pixels edge align into 2 704 pixels, and do not edge align with 720 pixels. This will cause a change in the sharpness of the picture, the extra pixels are put into the 352 picture by blurring them in. And it's going to show up on any player when you play back that 352 encode since it was permenantly put into the output by which way the encode was done.

That depends entirely on the resize method.

No, it doesn't. Edge alignment happens with 704 to 352, and not with 720 to 352, period, in a resize. You can sample outside of the actual pixel area while doing a resize, but then it's a filter not just a resize, even if someone calls it a resize. A weighted resize will have a distortion in the 720>352 that isn't in the 704>352. You can do a shortcut algorithm with the 704>352 resize, but it produces the same results as the weighted algorithm for that resize just faster.


And for those having difficulty seeing the reasoning here are some pictures:

http://home.nc.rr.com/alan69/resolution/

720 cropped to 704 resized to 352 has plain 2 into 1 distortion.

720 resized directly into 352 has an additional rolling distortion pattern on top of the 2>1 type distortion, which will be seen as additional softening in a normal picture. Exactly as I've said all along. Not dividing evenly always creates a shifting alignment error, and that's going to show up in video as extra blurring. Precisely the same softening as in an offset pole alignment motor vs one with repeating pole alignment motor, just in video instead of physical motion. I didn't even look for it until now, I knew the distortion was there from other experience with similar phenomena.

The extra blurring does get to happen at the same time in the 720>352 workspace, so it may actually be preferred as looking better in many cases. Just as the offset error is often accepted in a motor because you get smoother motion in return. But cartoons etc that suffer from loss of edge detail should look better with the crop first.

Alan

[junkmalle]

http://home.nc.rr.com/alan69/resolution/

Did you start with that resolution image that was posted above? If so your software simply subsampled (ie picked every other pixel of) the source image. The lower right hand circle contains alternating single pixel wide vertical black and white lines. Your "cropped to 704 resized to 352" image shows a pure white circle! If it had averaged together pairs of pixels it would have been medium gray. Not good examples of your basic argument (although I agree that a 2:1 reduction gives slightly better image quality).

[petar]

Alan69
I agree with junkmalle. Had you used, for example, bi-linear resampling the circle would have been grey in both cases, whether 704 to 352 or 720 to 352 (somewhat greyer for 704 to 352). Nearest Neighbour is the only algorithm that would have given you that outcome and that's a very fast but not a very good algorithm, so it's not a very good comparison (i.e. you wouldn't use it anyway in real resizing, so there's no point in using it in comparisons).

As I mentioned before, resampling creates a brand new image of the old one (as an analogy, a new empty image is overlaid on top of the old one and then drawn from scratch referening the old one). Resampling from 704 and from 720 to 352 will give you different images both from each other and from the original. It's difficult to compare which is better (because the colours will be different plus some parts will look better on the first one and some on the second one) however if you compare stuff like edges, angled lines, curves, etc. you'll see that there's very little (if any) difference.

Personally I'm not bothered whether cropping 720 to 704 is better for resizing to 352 (as I mentioned, if you resample with a good algorithm, it makes no difference) but whether the aspect ratio is preserved on all hardware players.

Can someone answer this for players that play SVCD, DivX, PS2 or Xbox? It wouldn't be too difficult to test. Create an MPEG video of 720, then crop to 704 and author both in two separate VTSs. When playing on a TV screen, put your finger on an object near the left or right edge and when the second video plays see if the object remained exactly where your finger is (there's only 2% difference so better if the object is a vertical line or something like that). If the same pixels are used the object shouldn't move. On a software player it will move. I'm just curious if the players I mentioned above act as the software players or as a normal DVD-Video only player.

[trevlac]

352 pixels edge align into 2 704 pixels, and do not edge align with 720 pixels. This will cause a change in the sharpness of the picture, the extra pixels are put into the 352 picture by blurring them in. And it's going to show up on any player when you play back that 352 encode since it was permenantly put into the output by which way the encode was done.

That depends entirely on the resize method.

This is absolutely correct. Take that resolution.png I posted. Open it in virtualdub, and try a bunch of resize (down size) methods with the resize filter.

Nearest neighbor will actually keep the most sharpness. This is because it just throws out pixels. This results in a distortion. Aliasing. From a quick test it looks like lanczos3 720->352 and 720->360 are the same. There is some aliasing (because only the 3MHz circle should be anything but grey). Bicubic, has more distortion (it seems). The evil bilinear actually keeps 'more detail' but this is really distortion. Precise bilinear blurs more (but this is more proper).

Anyway ... it's pick your poison. I like distorted detail over blur.

[trevlac]

No, it doesn't. Edge alignment happens with 704 to 352, and not with 720 to 352, period, in a resize. You can sample outside of the actual pixel area while doing a resize, but then it's a filter not just a resize, even if someone calls it a resize. A weighted resize will have a distortion in the 720>352 that isn't in the 704>352. You can do a shortcut algorithm with the 704>352 resize, but it produces the same results as the weighted algorithm for that resize just faster.


And for those having difficulty seeing the reasoning here are some pictures:

http://home.nc.rr.com/alan69/resolution/

Alan

Oh ... I see you already did some tests. Some comments .... ~ 3.3MHz is the max you can fit into a 352 frame. If you see anything but grey in 3 of the 4 circles, it is distortion. A proper resize is actually a resampling of the analog wave. To properly sample an analog wave, one should sample above 2x the highest frequency. So when you downsize ... you should blur the detail and then resample. This is actually the 'proper' method. However, the question is whether the distortion or the blur is more objectionable.

I'm with you and I don't like the blur.

So ... back to your point. Is a resize from 720->352 better, or one from 704->352 better?

The 1st skews the AR a tad. And it can not properly hold as much detail as the second (because it is trying to squeeze more into 352).

But the results can be vastly different depending upon the resize method. I tried Lanczos3 in VDub for both and really see no difference. Lanczos3 is the most technically proper method.


Edit
Yep ... as Junkmalle says ...

[trevlac]

Personally I'm not bothered whether cropping 720 to 704 is better for resizing to 352 (as I mentioned, if you resample with a good algorithm, it makes no difference) but whether the aspect ratio is preserved on all hardware players.

You know your resampling stuff.

Do you not believe the DVD Demystified quote I posted? It says the spec has the same rate for 704 and 720.

As far as what DVD players do ... who knows. TV's could also effect the AR. I looked at comming up with a good visual test for everyone .... but the difference is so small, I'm not sure people could tell.

Here is a pic from my test disc. Change it sigtly (2%) and see if you can tell without a reference. Maybe you have to compare a printout to see if you can tell on a TV. I thought about maybe a DVD with a subtitle overlay for compare. Not Sure.



PS: Arachnotron read this thread. And told me this ....

If you use my old test DVD's to check out how many pixels and lines
fall into the visible area of your TV you will probably find it is not
exactly 4:3 either. Most TV's are a few percent off too. Also, when I cap NTSC tapes, often the active width is not NTSC spec. Why? was the tape
already wrong AR? Did they crop? Without a reference in the scene you will never know.

Kinda puts the importance of the whole AR issue in perspective.

Here is his writeup on this with respect to capture .....
http://www.arachnotron.nl/videocap/site/capture_area2.html

[lordsmurf]

I remember shopping for tv's a couple years ago. You can visually see the screwed up AR on tv sets when you have 50 of them turned onto the same station side-by-side 4 racks tall. You'll now notice most stores don't have these kinds of displays.

[junkmalle]

Here's the 720 pixel image, cropped to 704, then resized to 352x240 with VirtualDub's bilinear resize filter (I know everyone's been talking about 352x480 but the result is similar horizontally and the lossless PNG file is under 50 KB at the smaller size):



Here's the result of resizing the 720 pixel image directly to 352x240 with the same bilinear filter:



And finally, 720x480 to 352x240 with VirtualDub's lanczos3 filter:



704x480 to 352x240 with lanczos3 looked similar.

In my opinion the last image is the most accurate. The source image is 50 percent black, 50 percent white (aside from the text). As the image is shrunk the small details should become a single shade of gray. The Lanczos3 filter did this pretty well and had the fewest moire (aliasing) problems.

I created an animation with the test pattern panning slowly across the screen and another with the pattern rotating. Both bilinear images had heavily strobing moire patterns. The lanczos3 images had them too, but not nearly as noticable.

[petar]

Do you not believe the DVD Demystified quote I posted? It says the spec has the same rate for 704 and 720.

As far as what DVD players do ... who knows. TV's could also effect the AR. I looked at comming up with a good visual test for everyone .... but the difference is so small, I'm not sure people could tell.

I do believe DVD Demystified (the site has been around since the very early days of DVD) but I'm suspicios when someone references the DVD specs (as I mentioned before, a respected web site can't just quote the DVD specs because there's a non-disclosure agreement if you want to obtain them; DVD Demystified also mentions this).

So I don't know if this applies to all players. Software players which resize the picture on playback will display a different picture for 704 and 720. Xbox and PS2 also resize so I would think they'd act as the software players. Players that can play more formats than DVD-Video are also capable of resizing. What if the DVD specs allow for same pixels for 704 and 720 to make it cheaper for the hardware manufactures but then also don't discard that you can use either resolution as full screen? Because I don't understand why software players would resize both 704 and 720 to 4:3/16:9 if the DVD specs say that both should use the same pixels. This would make the software players non-DVD compliant. And I can't believe that they were designed like this for simplicity. To imitate the hardware players all they have to do is resize 720 2% wider (for example, if they play 4:3 720x480 DVD video in 640x480, to play it in 653x480 instead).


As for testing, I know it's difficult to measure correctly the aspect ratio, especially when you take into consideration overscan, different TV technologies, etc. but it shouldn't be too difficult to tell if you have exactly the same picture when played in a sequence on a same TV set, first a 720 video and then the same but chopped to 704. And then as a reference a third where 720 is resized to 704 rather than chopped. If you play the three videos in a sequence you should be able to tell if something is changing.

[petar]

Here's also the same image resized with TMPGEnc (which I'd image most people using it to encode will use to resize as well) cropped to 704 and also as full 720. My far-from-perfect-eyes find it hard to tell which one is better.
(They are jpegs due to file size but when encoded to mpeg you'll have similar quality loss anyway; the bitmaps on my PC are also difficult to tell which is better).




As I mentioned before, don't compare stuff like colours because none is true to the original, but as far as bluring, edges, lines, etc. I find it hard to tell that one is better than the other.

[trevlac]

I do believe DVD Demystified (the site has been around since the very early days of DVD) but I'm suspicios when someone references the DVD specs (as I mentioned before, a respected web site can't just quote the DVD specs because there's a non-disclosure agreement if you want to obtain them; DVD Demystified also mentions this).

So I don't know if this applies to all players. Software players which resize the picture on playback will display a different picture for 704 and 720. Xbox and PS2 also resize so I would think they'd act as the software players. Players that can play more formats than DVD-Video are also capable of resizing. What if the DVD specs allow for same pixels for 704 and 720 to make it cheaper for the hardware manufactures but then also don't discard that you can use either resolution as full screen? Because I don't understand why software players would resize both 704 and 720 to 4:3/16:9 if the DVD specs say that both should use the same pixels.

I was actually referring to the Book by Jim Taylor. You can read it on Amazon using their 'search inside' feature. Page 292

So your logic goes ... Any reference that truely knows can not say. So anyone who says ... can not truely know. A conundrum indeed. I guess anyone without $5000 will never know.

It's easy for me to understand why software players don't work proplerly. They are developed by people who don't know video. They are notorious for not opperating properly (or maybe working good enough). Try some 720 subtitles on a 352 source some day.

PS: The 'same pixels sometimes' is an interesting theory. But I'd be disappointed if that were the case. Perhaps they say "use 720,704,352 ... ect. ... and do whatever you want to make them fit."

Please don't take any of this the wrong way. You certainly have a good understanding of how many things probably work. Reading over things I posted in the past, I don't even think I'm a reliable source. :P I certainly have not spent $5000 for the privlidge of reading a spec. Heck, I'm still trying to get thru the $10 I spent on the 9/11 commission report.

[trevlac]

Here's also the same image resized with TMPGEnc (which I'd image most people using it to encode will use to resize as well) cropped to 704 and also as full 720. My far-from-perfect-eyes find it hard to tell which one is better.

Looks like bicubic to me.


Since we are posting resolution pics ... I thought I'd post what VHS does to that chart. Since I'd suppoze most people don't muck with DVD source like this.

VHS has about a 3.2Mhz bandwidth. A 352 frame can hold just about this. So if you start with VHS, those details are just not there. Notice the vertical detail is. Something like VCD can only contain ~ 1/2 the vertical detail of VHS. VCD is bad.



edit
Woops!

I thought that was 11K not 110K. Bladrick you cheapskate! I'll host it.
/edit







Ivo, please don't fill up my inbox for saying stupid things like VHS fits in a 352 frame.

[NamPla]

I like following (or trying to ) Trevlac's comments on topics like this...

What I'm getting from this thread is that it DOESN'T MATTER whether you go 720-to-352, or 720-crop to 704-to-352!!!

Probably the first is better.

It may send the AR out of whack slightly (ON YOUR COMPUTER!), but on your TV set, IT REALLY DOESN'T MATTER. Because DVD players/TVs/etc do things to the AR anyway...

Is this correct?

[trevlac]

Fish Sauce
A good source of vitamin B and protein, fish sauce is made from -- surprise, surprise -- salted fish. In addition to being a valuable cooking ingredient, fish sauce also doubles as a dipping sauce for everything from vegetables and seafood, to spring rolls and more. Some Chinese brands call it "fish gravy." In Vietnam it is known as "nuoc mam" and in Thailand, "nam pla."

Yummy !


My opinion is ....

704 fits inside of a 720 frame. 352 is 1/2 of the 704 portion of the frame. If you had a 720x DVD, the most 'proper' way to make a 352 DVD would be to crop to 704 then resize. Lanczos3 is a higher quality resize and it shows less distortion than the other common methods. To the point that a starting point of 720 vs 704 does not matter.

However .... you probably can not tell the difference on something like VHS source as to what you do, as long as it is reasonable. (On a PC or a TV).

A use for this kind of information would be if you were making computer graphics that should overlay video in something like a dvd menu. IF you want it to line up ... you have to know how to change the sizes around.


PS:

If I have not made it clear ... I'm a shmoe who probably talks/reads about video more than he messes with it ... and definately messes with it more than he watches it.

[junkmalle]

I ran tests on my two DVD players. Starting with 720x480 image I made one MPEG of that frame size. Then I cropped down to 704 and made a file with that size.

On one of my DVD players (Yamakawa DVD 218) the two files played identically. So, effectively, that one either cropped the 720 down to 704, or padded the 704 up to 720 while playing.

On the other player (Liteon DVP 2002 MPEG4 player) the 704 image had less of the edges visible. Apparently that player either scaled the 704 up to 720, or the 720 down to 704.

Both players treated the 352 pixel wide images the same way they treated the 704 pixel wide images.

I suspect the DVD spec isn't too specific about this because the 2 percent aspect ratio difference is likely less than the typical AR/overscan error of most televisions.

[petar]

I ran tests on my two DVD players. Starting with 720x480 image I made one MPEG of that frame size. Then I cropped down to 704 and made a file with that size.

On one of my DVD players (Yamakawa DVD 218) the two files played identically. So, effectively, that one either cropped the 720 down to 704, or padded the 704 up to 720 while playing.

On the other player (Liteon DVP 2002 MPEG4 player) the 704 image had less of the edges visible. Apparently that player either scaled the 704 up to 720, or the 720 down to 704.

Both players treated the 352 pixel wide images the same way they treated the 704 pixel wide images.

I suspect the DVD spec isn't too specific about this because the 2 percent aspect ratio difference is likely less than the typical AR/overscan error of most televisions.

That's exactly the test I had in mind. Thanks junkmalle for doing it .

It is as I suspected then. A standard DVD-Video player will use the same pixels for 704 and 720. However the Liteon one, which by the name I assume plays MPEG4, acts like a software player because it has the capability to resize. I'd imagine PS2 and Xbox will do the same (anyone tried it on those?).

It's quite possible that it is as you say - because there's only 2% difference, the DVD specs may be left open to hardware manufacturers to use the same pixels for 704 and 720 in order to simplify the equipment (i.e. make it cheaper) although the specs may not be strict that all players must limit to this if they don't need to.

[NamPla]

PS:

If I have not made it clear ... I'm a shmoe who probably talks/reads about video more than he messes with it ... and definately messes with it more than he watches it.

Joe Shmoe (aka Trevlac) -- I enjoy your comments -- BUT --- you should watch more magnificent movies! ...and Yes, a good Thai stirfry goes well with those beers, pal...

Both players treated the 352 pixel wide images the same way they treated the 704 pixel wide images.

Junkmalle, didn't you just say your 2 DVD players treated 704 differently?

[Alan69]

Fish Sauce
A good source of vitamin B and protein, fish sauce is made from -- surprise, surprise -- salted fish. In addition to being a valuable cooking ingredient, fish sauce also doubles as a dipping sauce for everything from vegetables and seafood, to spring rolls and more. Some Chinese brands call it "fish gravy." In Vietnam it is known as "nuoc mam" and in Thailand, "nam pla."

Yummy !


My opinion is ....

704 fits inside of a 720 frame. 352 is 1/2 of the 704 portion of the frame. If you had a 720x DVD, the most 'proper' way to make a 352 DVD would be to crop to 704 then resize. Lanczos3 is a higher quality resize and it shows less distortion than the other common methods. To the point that a starting point of 720 vs 704 does not matter.

http://abcnews.go.com/wire/World/ap20040820_2082.html

LOL one of these ships was carrying fish sauce. Why does fish sauce make me think of fish-head soup or what's in the commode after you've had the runs? It's obviously decent or the Romans would not have been shipping it..


A 352 pixel fits over 2 704 pixels, and parst of 3 720 pixels. Blending them together by their percentage of included area is a weighted resize.

Note that ANY other function has additional filtering, not just a resize. If you look outside the destination pixel's area for hints to modify its value for what you think is better, then it's filtered. If you weight some of the interior included pixels more than others with non-linear math, then it's filtered.

Bicubic and Lancos3 are resizing filters. While they may even out the distortions and cause a more pleasing effect on many pictures, they are not 'more correct' or 'correcting' anything. They are adding an additional distortion, and it does not reduce things it only spreads out the other distortion. 352 cannot display the lower right circle, and it actually takes even more distortion to come up with the graying out vs the wider banding. It is coming up with it's gray by adding more blurriness. While it may seem more correct to some of you to end up with gray, note that this will add yet more softening on a normal picture, as well as adding more encoding time. It only seems more correct for that one case, over an entire random picture it will slightly worsen the picture.

As for usefulness, if all you ever did in your life was video, then you could simply try a lot of things out and see what you like. But if you understand what you're doing and the ideas behind them, then things become far more simple and useful, since you'll be able to narrow your options to a few instead of dozens for each situation by thinking about it a few seconds. And if you have any other interests, understanding useful ideas carries over and multiplies. These same ideas are there for aliasing and anti-aliasing, beat interference in audio, noise reduction, mechanical vibration and motion, and many other fields. As with everything else, 10% of advanced ideas do the work in solving 90% of advanced problems, and the general idea behind this is definitely in that 10%. So working it out can gain you much more than just the immediate advantage, and the longer you know them the easier it gets. Instead of being an advanced video conversation, this whole discussion is just a simple reapplication of old familiar ideas to video, which is just another subject. Wouldn't be this easy to me if I hadn't invested the time and used these ideas in several other subjects. It's also why I don't get bogged down with people interjecting useless info or saying I'm wrong. Used them for 20 years plus and worked out that they're a consequence of the math involved a long time ago, and I take no one's word over basic math, not even mine the once in a while I think out loud before doing the math. Once a mathematical pattern is established math is always right from that point on, and these are basic math effects from 704 being evenly divisible by 352 and 720 not. Other considerations like aspect ratio effects etc are totally irrelevant to this beat distortion, it's an effect of the numbers and will always be there. Even if you use some other filtering to try and hide it, it still exists.

Alan

[NamPla]

Alan69, that kinda makes sense...kinda...

But I warn you, if you base your life on mathematics too much you will freak out into quantum physics & black holes & singularities etc :P

DON'T GO THERE! (I warn you, I've been there, don't open that BOX!)

The thing that matters, of course, is the visual impression off of the TV set.

What do you think of Lanczos4Resize instead of Lanczos3?

(BTW the Romans were shipping cocaine, marijuana, beer, etc... not just fish sauce! )

I have tried both 720-to-352 and 720-crop704-to-352 and encoded/burnt to DVD. Watched on my telly. I CANNOT TELL ANY DIFFERENCE WATSOEVER!!!!!!!!!! Actually, 720-to-352 looks better, just because...

[junkmalle]

Both players treated the 352 pixel wide images the same way they treated the 704 pixel wide images.

Junkmalle, didn't you just say your 2 DVD players treated 704 differently?

Sorry if it wasn't clear: the Liteon handled the 352 pixel image the same way it had handled the 704 pixel image, and the Yamakawa handled the 352 pixel image the same way it had handled the 704 pixel image.