The importance of resolution?

[davidian]

Hello. Let's make some comparsions here:

I'm going to make a VCD and an SVCD from VHS source. I have an ATI all in wonder Radeon, and I'm going to capture to AVI with that looseless codec (huffhuyv or whatever it's called).

If I would make a VCD, how much would the quality differ, if I encoded all files with tmpeg, after capturing them at following frame sizes:

352x288, 352x576, 640x480, 704x576

I can understand, that capruting 704x576 and then encoding it to 352x288 with tmpeg would make better quality than capturing directly to 352x288 and encoding it.

But how much better? Is there A LOT of difference? Between all formats I stated?

And for making SVCD, is there big difference between image capturing frame sizes:

480x576, 640x480, 704x576

?

[iant]

I've found that when making a standard VCD, the difference between 352x288 and 704x576 is incredible. The best rule seems to be 'garbage in;garbage out'. By feeding in an AVI that is four times the resolution, you're giving the encoder much more information to work with. ( Yes, I know VHS doesn't have that high a resolution to start with but it still makes a big difference to the end result. )

If you want me to quantify the main difference I'd say that 'blocking' reduced probably 25%-50% by stepping up the capture size. The one drawback is that because the encoder has to resize it, capture time goes up. Can't win 'em all.

[niteflights]

From my understanding resizing in Virtualdub produces better quality VCDs than allowing the encoder (TMPGENC) to resize. Is this everyone's experience?

Niteflights

[deltaboy]

i agree. with a higher resolution, it seems easier to apply filters and such to better your AVI and also not producing too much bluring. i wouldnt use any filters in any MPEG encoding software because thats not its main purpose. doing them in a progy such as VDub and then doing a straight encode can give you maximum results.

what filters are you guys using in VDub? is there any i dont know of?

[Bullworth]

So far that's been mine. TMPGEnc does some weird things to the video file when resizing and the quality isn't as good as VirtualDub; probably because the resize filter provides precise bilinear and bicubic which are better at enlarging or shrinking images without distortion.
Contrary to what many other say here. I have found that capturing at 640x480 and then resizing to 352x240 gives me much better results than capturing straght to 352x240. When I have I used to get there strange artifacts around moving objects throghout the entire capture that I called macroblock auras. The only way to get rid of it was to set the noise filter as such levels that it would take forever to get rid of it.

[davidian]

...that's what I thought

But, what I had no idea of, was about the Vdub resizing vs tmpeg resizing!

What do you mean then? Capture to highest resolution possible, and then use vdub to frameserve with resize filter and then encode with tmpeg.. ?

What about the de-interlacing, should I use Vdub or tmpeg filter? Does it matter?

But when ripping DVD's I always use tmpeg for the resizing.. should I use vdub here too, to obtain best results?

[acerglenn]

Dumb Qusetion, hope someone can help me...

I may buy a card whose maximum capture resolution is 640 x 480 (as it's the best quality, cheapo that have AV inputs).

But, I want to make SVCDs which require a resolution of 480 x 576.

Will resize and eventual svcd work out all right if i capture at 640 x 480

thanks for you help!

[Just G]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-09-20 19:15:26, niteflights wrote:
From my understanding resizing in Virtualdub produces better quality VCDs than allowing the encoder (TMPGENC) to resize. Is this everyone's experience?</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

It's not really accurate to make a blanket statement about VirtualDub's resizing being inherently superior to TMPGEnc's. IIRC, TMPGEnc uses a bicubic resizing algorithm. With v1.4.7, VirtualDub allows you to choose from seven different algorithms.

• A precise bilinear will generally provide a higher quality reduction.
• The precise bicubics will generally provide a higher quality enlargement.
• VirtualDub's Nearest Neighbor is fast, but will provide a lower quality resize in either direction than TMPGEnc.

While both Precise bicubic and Precise bilinear sharpen images somewhat, bicubic does moreso than bilinear. When you are reducing an image, the added sharpening that comes with bicubic resizing looks good to the eye, but not so to the MPEG encoder.

So, if you want to reduce, then yes, setting VirtualDub to do a precise bilinear resize prior to feeding to TMPGEnc will generally produce better output.

Setting VirtualDub to do a nearest neighbor resize, however, will generally provide a lower quality output than having TMPGEnc do it.

Also, there can be a good deal of difference among the various algorithms when resizing live action vs. animation (solid color masses, sharp edges, etc.)

Concerning iant's comments on capturing at 704x576, I am curious as to the logic behind this statement - isolated to this thread regarding a VHS source. VHS has a resolution around 320x240 (NTSC, PAL is about 320x288), plus a few lines of overscan. Your VCD will wind up 352x288.

Capturing at a horizontal resolution of 704 lines from a 320-something source requires an on-the-fly real-time enlarging, which will only be subsequently reduced. It really doesn't make sense to do this. With video sources that have greater than 352 lines, DV, or DVD, for example, you would want to capture larger and reduce. But if the resolution isn't in the source itself, as in VHS, capturing at a greater resolution can't make quality magically appear.

The reduction of 'blocking' could very easily be attributed to the softening that can come with enlarging and subsequently reducing the video - load some 352x288 video into VirtualDub and chain two resize filters: first enlarging to 704x576, then reducing to 352x288 and see for yourself. All MPEG encoders dislike sharp edges; softening the image's edges will inevitably reduce blocking artifacts at the potential loss of clarity.

You will want to capture at a vertical resolution of 576, however; this is not in dispute. But, you shouldn't need to capture at a resolution greater than 352x576.

As you originally asked, davidian, you will get better VCD MPG's if you cap at 352x576 vs. 352x288. If you capture at 352x288, you will only be capturing the first field of each frame, whereas capturing at 352x576 will capture both fields. By capturing at 288, you will be losing motion quality, which is exactly the problems that Bullworth says he sees when capturing at 352x240(288). But you won't get higher quality by capping from VHS any larger than 352x576.

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>Capture to highest resolution possible, and then use vdub to frameserve with resize filter and then encode with tmpeg.. ? </BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>
Capture to the highest resolution necessary, then use VirtualDub's Precise bilinear resize filter and frameserve to TMPGEnc.

As for your follow-up regarding deinterlacing, I've always found that using a VirtualDub filter that has been designed to deal specifically with this task is more effective than TMPGEnc. Try Gunnar Thalin's deinterlacing filters home.bip.net/gunnart/video. You'll want to play with the settings and see what values your video demands. What you want to do ideally, is to eliminate as much of the interlacing lines as possible without "deinterlacing" lines that should stay. Set the Show deinterlaced areas only and preview how much is being deinterlaced. (Don't forget to turn this off, before starting the frameserver.)

Nutshell: When capturing from VHS, capture at 352x576, using HuffYUV at YUY2 and Predict median. Load cap's into VirtualDub. Add the deinterlacing filter. Add the resize filter. Resize to 352x288 with Precise bilinear. (Or, try Chris LaRosa's 2:1 Vertical Reduction filter - www.geocities.com/cplarosa/video/v21red.htm. It uses a slightly different algorithm than VirtualDub's resize and was created specifically to reduce 576 video to 288.) Start frameserver. Open the .vdr in TMPGEnc and all you should have to set therein is the Motion search precision to your liking. (Many have stated that the trade-off between High and Highest quality vs. encoding time is negligible. I tend to concur.)

Hope this helps.
Just G

[davidian]

Just G, thanks a lot for this explaination, it really helps! Very very good, I will try this soon.

But.. hmm. What about capturing from VHS and then making an SVCD.. = pointless then? Is the ideal format when capturing from VHS making an XVCD at 352x576 perhaps..?

[Just G]

<TABLE BORDER=0 ALIGN=CENTER WIDTH=85%><TR><TD><font size=-1>Quote:</font><HR size=1 color=black></TD></TR><TR><TD><FONT SIZE=-1><BLOCKQUOTE>
On 2001-09-21 08:10:28, davidian wrote:
Just G, thanks a lot for this explaination, it really helps! Very very good, I will try this soon.

But.. hmm. What about capturing from VHS and then making an SVCD.. = pointless then? Is the ideal format when capturing from VHS making an XVCD at 352x576 perhaps..?
</BLOCKQUOTE></FONT></TD></TR><TR><TD><HR size=1 color=black></TD></TR></TABLE>

Damn, took too long writing my reply - grabbed lunch between - and when I tried to submit it, it said I needed to log in, thus losing my message.

Anyway, SVCD might be an option. Really, you'd have to try it and see how it looks.

I've never tried XVCD and don't know how an MPEG-1 decoder would handle interlaced video at 576 horizontal lines.

I had the URL before, but lost it. Do a search on http://www.google.com for Nicky Pages. There's some good info on video capture including things like interlaced video.

The ideal format for VHS captures is a well made VCD - its frame size is near enough VHS's and it has a sufficient bitrate to contain all you'd need. An SVCD might turn out just as well, but you would have to do some enlarging which would depend on the quality of the source.

I can't recall some of the other things that I had written in my earlier draft which now floats in the ether somewhere, but I think these were my main points.

HTH,
Just G

<font size=-1>[ This Message was edited by: Just G on 2001-09-21 09:56:48 ]</font>

[jumperalex]

Just G

Ok so my vid card (old WinTV PCI) won't support 352x576. the closest it will grab is 352x480. so the question is then: will the fact that i'm not 2x the resized resolution (480/288=1.666) cause any artifacts when I use the resize filter down to 352x480? If so am I better of resizing down to 240? I ask this because of the comment you made about trying to capture both fields of each frame.

Also what about using the TV filter in VirDub?

Finally, based on your answers above ... Would it be worth then to go out and byu a vid card that DOES support 352x576.

To give perspective to my desires: I am looking to convert my 200+ VHS movie collection over to VCD (the SVCD vs VCD issue I was going ask about was already answered above; thanks)

Alex

[jumperalex]

PS What order should I put the filters in? And should I use the Interlace option in the resize filter or is that taken care of with the interlace filter. thanks in advance.

Alex

[davidian]

Well, Just G, what I meant was that when capturing from VHS, there is no point in capturing at greater resolution than 352x576, and since SVCD resolution is 480x576, I would have to enlarge, which would give me a bigger file and (probably?) not any better quality than 352.

But if I made an XSVCD (I'm not too sure about the expressions here , I mean .. an mpeg-2 video file with the resolution of 352x576, that'd be .. good .. if my DVD can play it, right..?

Or isn't it sure the quality would be better? mpeg-1 352x576 or mpeg-2 352x576 (and a higher bitrate I suppose).