Calibrating luminance levels/color for capture from VHS

[jagabo]

No wonder Cher's eyeballs look pale yellow.

With the low chroma resolution of VHS (about 40 lines across the entire width of the screen) you can't use small objects like that to judge color. Even if the picture had otherwise perfect colors small white objects would take on a color cast of the surroundings.

Three basic "colors" you should to know about:
black (RGB 0 0 0, or 16 16 16 in Rec601)
white (RGB 255 255 255, or 235 235 235 with much video)
gray (middle gray = RGB 128 128 128)

Since VirtualDub uses the standard rec.601 matrix to put YUV video onto the RGB screen, black should be 0,0,0, white 255,255,255.

The only time this would be not be true is if you used a PC.601 matrix to convert to RGB before giving the video to VirtualDub (ConvertToRGB(matrix="PC.601") in AviSynth. Then black should be 16,16,16, white 235,235,235. And you have to have your encoder set to use the PC.601 matrix when it encodes the video (assuming a YUV encoder) unless you manually adjust the range to 0-255 yourself in VirtualDub.

[sanlyn]

It looks to be letterboxed on the tape itself so maybe that's how it was broadcast or maybe the fact that it was broadcast with PAL dimensions is the issue?

Ooh, ahhh *waves hand in the air* - I have an explanation.

If the programme was made after 1998/1999 it's likely to have been produced in 16:9 SD, then formatted to 14:9 within a 4:3 frame. You want to keep the small bars at the top/bottom of the frame and it should be kept as 4:3.

If you're likely to watch this on a 16:9 TV, you could also crop the black bars off the top/bottom and generate thicker bars at the sides to create a 16:9 frame with the slightly narrower 14:9 image within it.

I see. BTW, if your DVD outputs 4:3, your TV will add pillars, etc. If you add your own 16:9 borders, it will show up on a 4:3 TV as letterbox-within-letterbox.

[sanlyn]

Another thing: I have...I have "sharpen" set to zero in my proc amp. I figured this is something that should be done in post-processing?

Usually, sharpen later -- after denoising. If you sharpen earlier, you sharpen noise and artifacts. Makes denoising rather rugged.

[sanlyn]

Another MPG of the earlier StillCher2. This time I used MCTemporalDenoise ("MCT") in YV12 before converting to RGB and correcting color in VirtualDub. MCT made color look cleaner (less botchy) and simpler to correct. Highlights are smoother, not as blocky as with VDub's 2D filter. Didn't need as much blue to clear reddish blotches. Still some noise in that left-hand red area; better learn to tweak some MCT elements to fix that. The "MCT" MPG has cleaner color: http://dc417.4shared.com/download/4i0Rhn5f/StillCher2_MCT.mpg.


[Attachment 8887 - Click to enlarge]

Attached is a ZIP file of MCTemporalDenoise (MCT) and its components, and some docs. The file unzips into 3 folders:

- "plugins". Copy all the scripts and dll's into AviSynth's plugins folder.

- "FFTW3DLL_copy_to_Win_System32". Copy the 2 dll's into Windows "System32". FFTW3.dll is used by FFT3D. I've always kept FFT3W and the older libfftw3f-3.dll packed together and in System32. I think libfftw3f-3.dll is still required by older plugins. If someone knows better about libfftw3f-3.dll, please let me know and I'll correct. I haven't had problems with it. More info about FFT3W at http://www.fftw.org/.

- "docs". Documentation that came with the plugins. Enjoy.

If anyone wants to check these contents, please do. I'm still learning to use this script.

[Cherbette]

Wow...quite a bit to look over. Since you fellas are more familiar with Avisynth (than I am thus far) you would typically do this sort of color tweaking with Avisynth rather than Virtualdub, correct? I need to find a tutorial somewhere on working with ColorYUV and it's adjustments. Lots here to catch up on...thanks again for all the feedback.

[intracube]

If you're likely to watch this on a 16:9 TV, you could also crop the black bars off the top/bottom and generate thicker bars at the sides to create a 16:9 frame with the slightly narrower 14:9 image within it.

I see. BTW, if your DVD outputs 4:3, your TV will add pillars, etc. If you add your own 16:9 borders, it will show up on a 4:3 TV as letterbox-within-letterbox.

Yes, like this:

- letterbox bars removed and formatted to 16:9 (a), and to the right how it would look on a 4:3 display
- letterbox bars retained (b) and how it'd look on a 16:9 TV*

*16:9 TVs can be set to expand/zoom the image so it looks like (a) whereas 4:3 TVs usually can't expand the image.

It would probably be best to leave it as is unless the footage is going to be edited with other 16:9 content.

[sanlyn]

Wow...quite a bit to look over. Since you fellas are more familiar with Avisynth (than I am thus far) you would typically do this sort of color tweaking with Avisynth rather than Virtualdub, correct? I need to find a tutorial somewhere on working with ColorYUV and it's adjustments. Lots here to catch up on...thanks again for all the feedback.

Indeed .

I've been using this popular default statement:

MCTemporalDenoise(settings="low", edgeclean=true, ecrad=3, stabilize=true, maxr=2)

Make sure you're in YV12. You can also try making "Settings" = "medium". Jagabo and some other members know far more about MCT than I do so far.

[jagabo]

A word of warning: McTemporalDenoise() is very slow, like most of the other filters that use motion compensation.

[sanlyn]

Nice pics, intracube. Thanks.

[sanlyn]

Since you fellas are more familiar with Avisynth (than I am thus far) you would typically do this sort of color tweaking with Avisynth rather than Virtualdub, correct?

Umm, not really. You'd generally want to work in YUV first, then RGB if needed. Avoid colorspace changes (sometimes you can't). I fine tune in RGB, if I don't need emergency fixes in YUV. VHS is just too "dirty" not to need both methods. That's why Adobe and others work in RGB. No way to localize color areas in YUV the way you can with curves and levels in RGB. I have godawful videos that could be fixed only with NeatVideo (RGB). IMHO one should learn to learn to work with both tools. Each has its strong points.

[Cherbette]

If you're likely to watch this on a 16:9 TV, you could also crop the black bars off the top/bottom and generate thicker bars at the sides to create a 16:9 frame with the slightly narrower 14:9 image within it.

I see. BTW, if your DVD outputs 4:3, your TV will add pillars, etc. If you add your own 16:9 borders, it will show up on a 4:3 TV as letterbox-within-letterbox.

Yes, like this:

[Attachment 8889 - Click to enlarge]
- letterbox bars removed and formatted to 16:9 (a), and to the right how it would look on a 4:3 display
- letterbox bars retained (b) and how it'd look on a 16:9 TV*

*16:9 TVs can be set to expand/zoom the image so it looks like (a) whereas 4:3 TVs usually can't expand the image.

It would probably be best to leave it as is unless the footage is going to be edited with other 16:9 content.

I have a 16:9 TV and that's what I'll be playing all of these on either streaming from the computer or burned onto a standard DVD.

[jagabo]

You have to be careful when resizing interlaced video.

[Cherbette]

Yeah I really don't wanna resize anything and damage the original aspect ratio of what's on the tape at all. I hate it when someone has attempted to "restore" a clip and they've made it widescreen but it just looks like a stretched mess.

[intracube]

You have to be careful when resizing interlaced video.

Yeah I really don't wanna resize anything and damage the original aspect ratio of what's on the tape at all. I hate it when someone has attempted to "restore" a clip and they've made it widescreen but it just looks like a stretched mess.

Done carefully, the aspect ratio and interlacing can be preserved.

But If you're happy to use the zoom function on your TV remote, you can avoid the extra work of cropping and re-formatting.

[sanlyn]

Yeah I really don't wanna resize anything and damage the original aspect ratio of what's on the tape at all.

Aspect ratio isn't the problem with resizing interlaced video, especially VHS. Low resolution is a terrific problem in itself, but resized interlaced never seems to go back together properly. Screws up color and detail too (what "detail" there is in VHS, and a lot of what you think is VHS "detail" is noise). Interpolation from old pixels and creating new ones -- it's a mess.

[Cherbette]

Ok here's what I've got so far:
http://www.mediafire.com/?bamu0ggvwjqlb2n

Using this script:
AviSource("StillCher2.avi")
ConvertToYV12
MCTemporalDenoise(settings="medium", edgeclean=true, ecrad=3, stabilize=true, maxr=2, interlaced=true)
Crop(22,40,-4,-24)
AddBorders(16, 32, 10, 32)

and adjust the Levels, HSV, and RGB values similar to what Jagabo posted. Still reading-up on the curves you posted San

So are all of the settings I posted screencaps of from the DVD-R passthrough accurate? Wouldn't change anything?

[Cherbette]

Also...for down the line once I'm done filtering the lossless AVI what should I use to encode to MPEG2? TMPGEnc Authoring Works?

[sanlyn]

Also...for down the line once I'm done filtering the lossless AVI what should I use to encode to MPEG2? TMPGEnc Authoring Works?

There are free encoders around, but kinda touchy to learn. TMPGenc Authoring Works isn't as able an encoder as the old classic (the new one has motion noise problems), but does lots of conversions. Check this one: still popular, easy to use, excellent color controls, and customers raised hell when TMPGenc tried to stop selling it: TMPGenc Plus http://tmpgenc.pegasys-inc.com/en/product/tp.html . $37. Free trial. You'll hear plenty of hot debate about it, too. Remarks here https://www.videohelp.com/tools/TMPGEnc-Plus. Awards mentioned here: http://tmpgenc.pegasys-inc.com/en/product/tp.html .A little slow (good work takes time. MCTemporalDenoise and NeatVideo are slow, too, and so is Premiere). Don't use their silly old wizard, stay with the real interface. A good guide here: http://www.digitalfaq.com/guides/video/convert-tmpg.htm .

Someone's bound to kick up some dust as soon as they see this post.

[Cherbette]

Thanks San...I got the one you recommended. I've used Tmpgenc before so this part should be easy.

[jagabo]

I'd use HcEnc via HcGUI.

[sanlyn]

And HcEnc is pretty good, too.

[Cherbette]

Any comparisons? The perfectionist at heart that I am wants the best of the best. Of course I'm sure that is subject to opinion.

[sanlyn]

TMPGenc and HcEnc will both work well. I like the image controls in TMPGenc, they've saved many horrible scenes for me. BTW your DVD settings look OK to me, similar to what I've used on my Toshiba.

Your AviSynthScript(3) video has good color, but some TV's might choke on those dense shadows. The white title is pretty hot, too (RGB 245). Looks good otherwise and tape noise is way down.

[Cherbette]

video has good color, but some TV's might choke on those dense shadows. The white title is pretty hot, too (RGB 255 and climbing up the right-hand wall)..

How would I correct this in my script? Also...if I were to correct the color using Avisynth rather than the RGB tools in VirtualDub, what should apply to my script? I assume that would be ColorYUV?

[jagabo]

You could crank the gamma up a bit to bring out more details in the shadows. You'll probably have to adjust the black and white levels if you do that. The "Dr John" titles look fine to me. Titles like that are often at IRE 100.

Gamma and black level adjustment of the latest video:

AviSource("New AviSynth Script (3).avi", pixel_type="YUY2")
StackHorizontal(last, ColorYUV(last, off_y=-11, gamma_y=80))
TurnRight().Histogram().TurnLeft()

You should make the gamma adjustment when you do the rest of your levels adjustment.

[Cherbette]

So then my script would read like this?

AviSource("StillCher2.avi")
ConvertToYV12
StackHorizontal(last, ColorYUV(last, off_y=-11, gamma_y=80))
TurnRight().Histogram().TurnLeft()
MCTemporalDenoise(settings="medium", edgeclean=true, ecrad=3, stabilize=true, maxr=2, interlaced=true)
Crop(22,40,-4,-24)
AddBorders(16, 32, 10, 32)

What could I do to correct the colors using Avisynth rather than the RGB methods I was using in VirtualDub?

[jagabo]

So then my script would read like this?

AviSource("StillCher2.avi")
ConvertToYV12
StackHorizontal(last, ColorYUV(last, off_y=-11, gamma_y=80))
TurnRight().Histogram().TurnLeft()
MCTemporalDenoise(settings="medium", edgeclean=true, ecrad=3, stabilize=true, maxr=2, interlaced=true)
Crop(22,40,-4,-24)
AddBorders(16, 32, 10, 32)

No, the StackHorizontal() and Histogram() stuff is just to show the before and after results and the levels graph. You should perform the gamma adjustment when you do the levels adjustment from your original source.

AviSource("StillCher2.avi")
ConvertToYV12()
ColorYUV(last, off_y=-11, gamma_y=80) #any other YUV levels/color adjustments go here
MCTemporalDenoise(settings="medium", edgeclean=true, ecrad=3, stabilize=true, maxr=2, interlaced=true)
Crop(22,40,-4,-24)
AddBorders(16, 32, 10, 32)

What could I do to correct the colors using Avisynth rather than the RGB methods I was using in VirtualDub?

I've lost track of what you have done so far. So all you did to the source was the MCTD and the border adjustments?

[sanlyn]

According to Cherbette's script and notes posted earlier, he cropped and used MCTD in AviSynth, then adjusted color etc. with the VDub Levels, HSV, and RGB filters. I think the low end of the Levels filter might have been set too far to the right, which brings the darks and upper midrange down closer to RGB 0. It wouldn't make sense to fix gamma and levels in AviSynth, then undo the work in VirtualDub. For the initial adjustment, I'd stick with one or the other, not both.

My histogram and CSamp says the white titles hit 255 consistently. But I see this on broadcast video all the time, doesn't seem to be that much of a problem.

[jagabo]

According to Cherbette's script and notes posted earlier, he cropped and used MCTD in AviSynth, then adjusted color etc. with the VDub Levels, HSV, and RGB filters. I think the low end of the Levels filter might have been set too far to the right, which brings the darks and upper midrange down closer to RGB 0.

Blacks should be at RGB=0 in VirtualDub. Near blacks can be lightened in many was as we've both described.

It wouldn't make sense to fix gamma and levels in AviSynth, then undo the work in VirtualDub. For the initial adjustment, I'd stick with one or the other, not both.

Yes, of course.

[Cherbette]

According to Cherbette's script and notes posted earlier, he cropped and used MCTD in AviSynth, then adjusted color etc. with the VDub Levels, HSV, and RGB filters.

That is exactly what I did But I remembered jagabo had stated I would probably get better results working with AviSynth rather than RGB in VirtualDub...so I'd like to try that way too.