Should I use 640x480 or 320x480 with cubic for VHS capture

[groove365]

I'm using VirtualDub capturing to XVID Quantizer 2 or 3 with a WinTV Go card.
I have a lot of tapes to back up so I would like to keep it around 1GB per hour if possible. If not I'll make them bigger.

So far in my tests the 320x480 with cubic looks better than the 640x480. The larger files have much more noise and are much bigger.

I still have a CRT 24'' TV.
I'm getting an HDTV soon probably around 40' and wonder if the bigger and better TV will make the smaller files look a lot worse.
The tapes are work related instructional videos and some old stand up comedy videos so I can live with some quality loss but I still want them to look good.

I need these files to play on a standalone DVD/Divx player.

Thanks everybody.

[jagabo]

With target quantizer encoding both should look about the same, but the bigger frame should give you a bigger file. At 1 GB/hr, ~2000 kbps, 320x480 or 352x480 will look better than 640x480 -- even on a big screen HDTV. I'd use a precise bilinear filter rather than bicubic -- to get a little noise reduction instead of noise enhancement.

Be sure your Divx/DVD player supports PAR/DAR flags so the picture comes out 4:3 at playback.

[groove365]

@jagabo
It seems that the vertical reduction cleans up some of the noise, that's why I said the 640x480 tests I've run haven't looked as good as the 320 and 352x480 tests.

When you say use a precise bilinear filter are you talking about the vertical reduction option in Virtualdub or an actual filter?
It's not really clear. I've been using the vertical reduction options.
If it's a filter where do I find it. It's not in my filter list.

I've been doing a lot of reading on this site while learning to use Virtualdub the past couple of weeks.
I want to say thanks for all the informative posts and clear explanations that you've written on video capture.
They've really helped me to learn what to do quickly.

I have one more issue I want to ask you about.

This past weekend I was testing CBR then I read some of your posts recommending target quantizer encoding.
I've noticed more artifacts in the tests I've run today using target quantizer encoding of Q3 and 4 (the Q2 was about 2GB per hour) than I have using CBR of 1400, 1600 and 2000.
In the target quantizer tests I've noticed a churning effect in the backgrounds and slight halos and churning around motion that CBR test don't have even at 1400.
Am I doing something wrong or is this normal.
I'm not seeing the advantages of quantizer encoding if the same size file has more artifacts.

Can you please provide a little clarity on whats going on.

Thanks for all of your help.

[jagabo]

It seems that the vertical reduction cleans up some of the noise, that's why I said the 640x480 tests I've run haven't looked as good as the 320 and 352x480 tests.

When you say use a precise bilinear filter are you talking about the vertical reduction option in Virtualdub or an actual filter?

Yes, I'm talking about the Precise Bilinear option of the Resize filter in VirtualDub. (Equivalent to BilinearResize() in AviSynth).

Note that reducing the frame from 640x480 to 320x480 is a horizontal reduction, not a vertical reduction. And that VHS has a horizontal resolution of about 350 pixels, so you're not loosing much by using 352 or 320 pixels rather than 640.

This past weekend I was testing CBR then I read some of your posts recommending target quantizer encoding.
I've noticed more artifacts in the tests I've run today using target quantizer encoding of Q3 and 4 (the Q2 was about 2GB per hour) than I have using CBR of 1400, 1600 and 2000.
In the target quantizer tests I've noticed a churning effect in the backgrounds and slight halos and churning around motion that CBR test don't have even at 1400.
Am I doing something wrong or is this normal.
I'm not seeing the advantages of quantizer encoding if the same size file has more artifacts.

Can you please provide a little clarity on whats going on.

Constant (Target) Quantizer encoding encodes every frame with the same quality. Constant bitrate encodes every frame with the same number of bits. (It's not literally every frame -- B frames are encoded with lower quality/bitrate than I and P frames, but we will ignore that for the time being since both methods can use B frames.) So let's say you encode a video at Target Quantizer 4 (because we know that will generate a lot of macroblock artifacts) and find the resulting video has an average bitrate of 1500 kbps. Then you encode again using a constant bitrate of 1500 kbps. Then you compare the two files...

What you'll find is that the "still" shots look better in the CBR encode, but the "action" shots look better in the CQ encode. To simplify matters let's assume the first half of the video is all still shots, and the second half is all action shots. And just to put some numbers on it let's say the first half of the CQ encode used 1000 kbps and the second half used 2000 kbps (so the average is 1500 kbps). It should be obvious now the first half of the video will look better in the CBR encode because it is encoded at 1500 kbps rather than 1000 kbps. And the second half of the video will look better in the CQ encode because it is encoded at 2000 kbps rather than 1500 kbps. So, although you may be seeing some portions of the video look better in the CBR encode other portions will look worse. Since most video consists mostly of still shots with occasional bursts of action it makes sense to "steal" a little bitrate from the still shots (reducing their quality by a very small amount) to give to the action shots (increasing their quality by a large amount). So, in general, the overall quality of the video should be better in a CQ encode compared to a CBR encode.

I would never use a target quantizer of more than 3. (Well, there are some exceptions -- like when I want to show something unrelated to the encoding quality but want a very small file to upload, or when I want to show examples of macroblock artifacts.) For "important" stuff I use Q=2, for day to day watching Q=3.

Note that "still" and "action" don't have their conventional meanings when used in regard to video encoding. The largest bitrate savings in video encoding come from not including the entire image at every frame. When parts of the image are the same as the last frame the encoder can just say "leave all those parts the same, just make these changes..." Or with motion vectors the encoder can say "move this block of pixels to this new location." For example, in a panning shot the encoder can say "shift the whole image left by 4 pixels, then fill in the right edge with this..." Anything that prevents those two techniques from working is "action". That includes things like static noise, fade-ins and fade-outs, strobe lights, complex motions of things like smoke, fire, heavy rain, reflections off of wavy water, etc.

VHS is a very noisy medium so it makes it much harder on the encoder to find still shots from which to shift bitrate to the action shots.

[groove365]

I ran some more tests and the resize bilinear filter does look better (less noise) than the Virtualdub resize option (once I figured out that I shouldn't use the interlace option).
Thanks for the detailed explanation of CBR vs quantize.
I'm going to stick to quantize 2 and live with the larger files.
I'd rather do it right the first time than have to do it again later because some things don't turn out right.

Thanks for all your help.