Capturing 704x480 --> Resizing to 640x480 == better quality?

[kisk]

I've heard you will yield better results if you capture tv at 704x480 then resize to 640x480, as opposed to just capturing at 640x480. Can anybody support this idea and if so can you tell me what resizing filters/settings work best (virtualdub's resize filter/divx's resizer, etc).

Let me say I'm capturing ntsc television and will be encoding it to divx 5.1.1 to be played back onto the television thru a standalone divx player.

Any additional comments/suggestions would be nice.

Thx! :D

kisk

[Sulik]

The answer is that it depends what capture card you're using.
With a decent capture card, it will usually be better to capture directly at 640x480, as it will be able to do a better job of scaling, since it will be performed by the ADC.

Scaling from 704 to 640 in software with simple bilinear or bicubic scaling will usually result in softening and/or asialing.

[Wilbert]

With a decent capture card, it will usually be better to capture directly at 640x480, as it will be able to do a better job of scaling, since it will be performed by the ADC.

which is false. See:

http://forum.doom9.org/showthread.php?s=&postid=433515#post433515

If you capture at 704x480 (filter and resize to 640x480) it might be easier to denoise your clip.

[trevlac]

@Wilbert

I don't think i4004's test answer the question.

He caps low and resizes up. He does not compare cap high, resize lower to just cap lower. PC 4:3 PAL vs NTSC is different this way.

Here are some more tests. i4004 of course disputes that the BT resizer is any good, but does not dispute that it always resizes. If you cap at your intended size you only get 1 resize. Come to think of it, i4004 disputes just about anything you say to him. ;)

I will gladly run more for you.

http://trevlac.us/wilbert/captest.htm
http://trevlac.us/wilbert/captest02.htm

I think kisk should do a test and tell us which is best for his/her setup. My guess is that there is no practical difference. As far as it being better for denoising, I don't know. I'm just a novice.

Trev

edit

"Scaling from 704 to 640 in software with simple bilinear or bicubic scaling will usually result in softening and/or asialing."

This does not have to be true. It depends on how you resize, and/or how your chip resizes. Do a test and see what works for you.

[Wilbert]

He caps low and resizes up. He does not compare cap high, resize lower to just cap lower.

http://forum.doom9.org/showthread.php?s=&postid=434727#post434727

Yes, he does. He compares the following:

1a) cap low at 368x576 [enlarged to 768x576]
1b) cap low at 352x576 [enlarged to 768x576]

2) cap high at 768x576 resized to 352x576 (avs: pointresize) [enlarged back to 768x576]

He just enlarged them _back_ again, so that the differences are more clearly visible.

Come to think of it, i4004 disputes just about anything you say to him. :)
:)

edit: strange your captures of 712 and 368 look better as 640. Why are the ones at 712 and 640 so different?

edit2: "PC 4:3 PAL vs NTSC is different this way."
Yeah, I forgot about this. You should compare his pics at:

1) cap low at 400x576 [enlarged to 768x576]
2) cap high at 768x576 resized to 352x576 (avs: pointresize) [enlarged back to 768x576]
better) cap high at 768x576 resized to 400x576 (avs: pointresize) [enlarged back to 768x576]

Can't see much difference.

[trevlac]

@Wilbert

I will have to re-look at ivo's tests. I think the resize up muddies the issue. More resizes = more filtering. But I missed that the reason was to see the difference better. Too bad no one wants to do resolution charts. :)

I'm not sure what you are comparing on "strange your captures of 712 and 368 look better as 640. Why are the ones at 712 and 640 so different?"

Can't really compare across the tests. I used different jpeg compressions across the tests. Give me a test and I will run it.

PS: I don't mean to give mis-information. I think every step in the process matters. People should test their setup. The safe way is to cap high and stay high.

Trev

[Arachnotron]

these tests focus on the BT878. The Philips chips use a completely different scaling algo and also sometimes will let you cap higher then the BT (which does not mean the quality will be better). So checking your own device remains important, especially since a lot of filter options in these chips can be configured by the driver. So even for the same chip, different brands may give different results!
I could not find what kind of device Kisk is using :confused:

Tomorrow I will buy a s-VHS tape to make a nice reference source for testing my SAA7134. No NTSC unfortunately. I could take a DVD for that, but I am not so sure DVD represents a real world capping source. Maybe I'll have some results next week.

Would be nice for Ivo too, since he really is interested in info about other chips. :)
We (well, Ivo and Trevlac mostly) are still fighting this one out over at everwicked (http://virtualdub.everwicked.com/index.php?act=ST&f=6&t=5896&)

[Wilbert]

I'm not sure what you are comparing on "strange your captures of 712 and 368 look better as 640. Why are the ones at 712 and 640 so different?"

Can't really compare across the tests. I used different jpeg compressions across the tests.

I was looking at the following pics (the second has rainbow effects, color bleeding, etc.):

http://trevlac.us/wilbert/712Wilbert.jpg

http://trevlac.us/wilbert/640%20VHS.jpg

This first one is that a capture at 712 and resize to 368? What did the original 712x480 look like?

[trevlac]

Originally posted by Wilbert
I was looking at the following pics (the second has rainbow effects, color bleeding, etc.):

http://trevlac.us/wilbert/712Wilbert.jpg

http://trevlac.us/wilbert/640%20VHS.jpg

This first one is that a capture at 712 and resize to 368? What did the original 712x480 look like?

The 1st was a compare of TV caps at 368x480, 352x480, and 712x480->368x480 (lancoz3). Full test (http://trevlac.us/wilbert/captest.htm)

The 2nd is a 640 VHS cap. Full test (http://trevlac.us/wilbert/captest02.htm). In that test, if you compare the 720x480->640x480 vs the 640x480 direct cap, there appears to be less color bleed in the higher size. I will check the original AVIs and post some more shots. It may be that the BT does NOT do a good job with the resize of color. Note: All caps were s-video, so the color bleed is interesting. Have to look at that damb spec again to see what they do with their comb/notch filters.

----------------------
On another note: I did some caps 712x480 and 368x480 VHS and TV. I did not get a chance to look at them, but I posted the huffyuv avi files, so you can slice and dice how you likey. I also grabbed pics of the register settings in case we needed to see what was happening with the driver (the chip has built in sharpening you know). I forgot to post them, but I will when I'm at home.

http://trevlac.us/wilbert/captest03.htm

[Arachnotron]

Note: All caps were s-video, so the color bleed is interesting.

VHS will allways have some colorbleed. luma and chroma do overlap a bit on VHS tape, so the vcr has to do a combfilter step. Using s-video over composite just avoid going through a second combfiltering step, so quality is better.

composite:
composite-combYC-remodulateC-store-playback-remodulateC-joinYC-composite

s-video:
s-vid-remodulateC-store-playback-remodulateC-svideo

[edit] the lowered chroma resolution of VHS tape does not help either

[trevlac]

@arachnotron

I thought VHS was 'color under', and did not overlap.
http://www.ee.washington.edu/conselec/CE/kuhn/vcr/vcr23.gif

http://www.ee.washington.edu/conselec/CE/kuhn/vcr/vcr.htm

Can't say I know much about VCRs though. Hanging around here may force me to understand. :D

[Arachnotron]

Great link! I have been looking for something like that page for ages. I found another one i'll post later. From it I understood beta was colour under, but VHS almost but not quite.

I may have misunderstood it of-course. Or it may be wrong. Both happens a lot with info from the net. :D

[trevlac]

@wilbert, @arachnotron

The BT spec does say it uses a 2 tap 32 phase filter for chroma horizontal resize. So if you follow the way I think about it, that means you are using only 64 coefficients vs 192 on the luma.

It also says that the Y/C seperation is by-passed if you use s-video. You can also do monochrom. That drops the U * V (to the mid point which is zero or 128 depending on how you count).

Maybe I can cut the pin on an s-video cable to test what Arachnotron is saying......Would that be a good test? .... so many tests, so little sleep, so little to show for it ;)


On another note,

I did find some code that appears to do polyphase interpolation/decimation. I can't follow sh0dan's/phaeron's stuff.
All the work is done in about 15 lines. It also appears that it only filters once (aka it does interpolation & decimation & filter in 1 step).

http://trevlac.us/code_lib/FIfConvSI.c.txt

PS: I hope it's ok to show here. There are libraries of this stuff available here (http://www.bdti.com/faq/1.htm)

[Arachnotron]

@Trev
I found another link from a company that makes VCR testing equipment.

http://www.sencore.com/custsup/pdf/TT189.pdf

It also states there is a slight overlap between the two bands.

Funny aside: according to the datasheet Philips leaves the combfilter on even for s-video. since it would not degrade the signal in that case. I'm not sure if that is true....

[trevlac]

@Wilbert

I did some more tests. I uploaded bmps so there is no effect due to jpeg.

http://trevlac.us/wilbert/712Wilbert.jpg (old one which is 712 resized to 368)
You asked for the original 712Wilbert that the above was based upon.

http://trevlac.us/wilbert/712Wilbert.bmp This is the original frame from the avi.

Also, I matched 2 frames of VHS caps. 1 from a BT resize to 640 and one from a 720 BT cap resized via lanczos. The color bleed is different, but I can't say the lanczos resize clean'd up the problem.

http://trevlac.us/wilbert/640%20VHS.bmp <<== BT resize
http://trevlac.us/wilbert/720%20lancoz%20to%20640%20VHS.bmp <<=== VDub resize
http://trevlac.us/wilbert/720%20VHS.bmp <<== Full 720 before VDub resize

As a note: Please let me know if you don't want the avi files. I'd like to remove them.

Trev
edit those damb spaces....

[Wilbert]

http://trevlac.us/wilbert/712Wilbert.bmp This is the original frame from the avi.
Is this a vhs cap, or directly from tv?

As a note: Please let me know if you don't want the avi files. I'd like to remove them.
You can remove them.

[trevlac]

Originally posted by Wilbert
Is this a vhs cap, or directly from tv?


TV ... Not a bad source huh ?


My intent on the tests was to see if color bleed was cleaned up if I cap'd high and resized after. (Vs cap low.) I did not see this in my test, although it seems reasonable that this may be an advantage of cap'in high.