Capturing from old VHS and improving quality of cartoon

[Abbadon]

Hello people,

I was wondering if you could give me your opinion regarding an old cartoon I am trying to recover from an old VHS tape. The recording was done with a standard VHS machine RCA using ChrisTV, the colors were adjusted manually using screen captures from a blog.

I am basically trying to make it look good, any suggestions are welcome.

[jagabo]

I think there's a little too much contrast. Blacks are too black, brights to bright. You should upload a sample capture, unprocessed YUY2 or YV12.

[Abbadon]

Hi,

Thanks for your time, the capture was done using the Huffyuv v2.1.1 codec, here is a sample:

http://www.mediafire.com/?01zsux3xnsru3qc

The password for this rar is password.

[jagabo]

That video is in really poor shape. It has terrible blending problems (many fields are a mix of two different pictures). Is the whole video like that? Can you get a cleaner capture? This is often a result of PAL/NTSC conversions. It can usually be fixed to some extent with SRestore() but it's much worse than normal here.

I initially thought there was a problem with the frame order. When the trap door opens it does so with a jerky up and down motion. But I think that was intentional on the part of the cartoonists.

The levels (contrast) are way off scale the darks are too dark and the brights are too bright. That leads to a loss of details in dark and light areas. Can you adjust the brightness and contrast while capturing?

Beyond that it has the usual problems you expect from VHS capture: lots of chroma noise (the purple and green splotches everywhere), lots of luma noise, over sharpening halos, and horizontal time base jitter. That last item can really only be fixed by using a quality s-vhs deck with a line time base corrector. You're probably not will to spend a few hundred dollars to fix that. If your VHS deck has a sharpness adjustment, turn it down. The same goes for the capture software. You can get better sharpening in AviSynth so it's best to avoid oversharpening by the VHS deck or capture software. I would also avoid noise reduction in the VHS deck and capture software (it is usually over aggressive and gives blending artifacts).

The sample images you posted don't have most of those problems. Are those from the actual video capture? What did you process them with?

[Abbadon]

That video is in really poor shape. It has terrible blending problems (many fields are a mix of two different pictures). Is the whole video like that? Can you get a cleaner capture?

The whole video looks horrible without any tweaking at the viewing application level.

The levels (contrast) are way off scale the darks are too dark and the brights are too bright. That leads to a loss of details in dark and light areas. Can you adjust the brightness and contrast while capturing?

Yes, I actually did it before capturing directly in Huffyuv v2.1.1 codec format, so what you downloaded was a video captured with some adjustments done manually using the naked eye.

Beyond that it has the usual problems you expect from VHS capture: lots of chroma noise (the purple and green splotches everywhere), lots of luma noise, over sharpening halos, and horizontal time base jitter. That last item can really only be fixed by using a quality s-vhs deck with a line time base corrector. You're probably not will to spend a few hundred dollars to fix that. If your VHS deck has a sharpness adjustment, turn it down. The same goes for the capture software. You can get better sharpening in AviSynth so it's best to avoid oversharpening by the VHS deck or capture software. I would also avoid noise reduction in the VHS deck and capture software (it is usually over aggressive and gives blending artifacts).

Sorry, I have no plans to purchase an expensive VHS deck, I would if I could, but this is just a hobby, I am currently using an old RCA VR546 4-Head VHS VCR.

The sample images you posted don't have most of those problems. Are those from the actual video capture? What did you process them with?

The images posted are the final result after processing the capture using VirtualDub and Neat Video, the final step is to encode the cartoon into mpeg2 for DVD burning.


Thanks for your recommendations, I guess I will have to accept what I have accomplished so far, due to work I can't learn AviSynth at the moment.

[redwudz]

This is a quick mod of the first JPEG using Virtualdub Mod and the Color Mill 2 filter. Probably better done with the actual video, but it might give you some ideas what is possible. http://www.infognition.com/cgi/getfilter?id=193&a=mirror

[Abbadon]

Wow, thanks redwudz. I didn't know Color Mill 2 existed, I will give it a try when I return home today.

[jagabo]

I wanted to cover all the things you can do before you start filtering. Capturing with better levels is important. Here's a video scope graph of the luma (brightness) of the AVI file:



The graph at the bottom indicates the brightness of the pixels above it. Dark pixels are at the bottom of the graph, bright pixels at the top. The luma channel in YUV video ranges from 0 to 255. But it should not have anything below Y=16, or above Y=235. When converted to RGB for display the darks below 16 will be crushed (all the same shade of black). The brights above 235 will be also be crushed. I marked Y=16 with a yellow line at the bottom of the graph, Y=235 at the top.

We can adjust the levels to get them withing spec (I used ColorYUV(cont_y=-36, off_y=-2) in AviSynth):



As you can see, all the pixels are now between 16 and 235 (I didn't add the yellow lines in this image). There are now more details visible in the dark areas of the picture (especially the trunk of the tree). But that solid white bar near the bottom of the graphs indicates the original video had lots of details even below Y=0. Those where crushed during the capture (there's no way of indicating values less than 0 in the YUY2 data so any parts of the picture that were darker than 0 became 0). There no way to restore any of those details with filters. If your capture card has a video proc amp you may be able to adjust it so that those darker areas aren't crushed. Then you would be able to see even more details in the dark areas.

[Abbadon]

This question will sound silly, but how can I apply ColorYUV(cont_y=-36, off_y=-2) to my current saved video?, Virtualdub reports an error saying that requires YUY input.

[jagabo]

You should be able to use an AviSynth script:

AviSource("filename.avi")
ColorYUV(cont_y=-36, off_y=-2)

In VirtualDub open that as if it's a video file. Ie, File -> Open Video File... select AVS script. If that is giving you the above error try this script instead:

AviSource("filename.avi")
ConvertToYUY2(interlaced=true)
ColorYUV(cont_y=-36, off_y=-2)

You can come close to the same result in VirtualDub alone but you have to be careful. If you apply the Brightness/Contrast filter first it will usually work because that filter can work with the video in YUV. If you apply some other filter before it, the video will usually be converted to RGB. That is where the crushing of the Y<16 and Y>235 happens.

If you enable Show Image Formats in the Video -> Filters dialog you will be able to see that the video is in YUV format for the Brightness/Contrast filter:

[Abbadon]

Hi,

Your script did the trick, the missing parameter was ConvertToYUY2(interlaced=true), this is how it looks from my end:

[jagabo]

It's odd that AviSource() didn't get YUY2 video directly from the HuffYUV decoder. Make sure the decoder wasn't set to output RGB. That would result in the darks and brights being crushed, the same as would happen in VirtualDub. You can test the colorspace with:

AviSource("filename.avi")
Info()

That will print information about the video on each frame:



The levels in your sample images aren't right. After conversion to RGB the blackest blacks should be 0,0,0, the brightest whites 255,255,255 (not in every frame of course, but over the entire video). There's a handy tool called CSamp that lets you see the RGB values of pixels on the screen as you move the mouse around.

[sanlyn]

Yes, I actually did it before capturing directly in Huffyuv v2.1.1 codec format, so what you downloaded was a video captured with some adjustments done manually using the naked eye.

It's impossible to judge out-of-range luma and color values by eye. You can read pixel RGB values directly off your Window dialogs, VirtualDub, and other viewers (but not media players) with a free (!) pixel reader like Csamp (http://www.netreach.net/~gavin/gavsfreeware/csamp.htm ). Pixel samplers are considered essential for use with Photoshop Pro, After Effects, Premiere Pro, etc. Csamp works as well as the big guys but doesn't cost $800.

Sorry, I have no plans to purchase an expensive VHS deck, I would if I could, but this is just a hobby, I am currently using an old RCA VR546 4-Head VHS VCR.

That's understandable. They aren't made any more, and finding one in good condition would be a gamble. What many do (and it works) is use a DVD recorder as a pass-thru line-level TBC. Feed your tape signal to the DVD recorder's input, but don't record directly to DVD (that's a no-no with noisy VHS). Instead, connect the recorder's output to your capture device. Panasonic, Toshiba, and most newer DVD recorders sold by Magnavox and Toshiba usually have built-in TBC's. I can't vouch for the lowest-end models. If you use a Panasonic, use only Line 1 input for TBC pass-thru.

The images posted are the final result after processing the capture using VirtualDub and Neat Video, the final step is to encode the cartoon into mpeg2 for DVD burning.

From your caps posted here, your Neat Video settings appear to be too high, along with their sharpener. If you're using Neat Video's default settings, go into the Advanced interface and learn to tweak them on your own. The defaults are too aggressive.

Thanks for your recommendations, I guess I will have to accept what I have accomplished so far, due to work I can't learn AviSynth at the moment.

Don't feel alone. You can do a lot with VirtualDub, used correctly.

[jagabo]

It's impossible to judge out-of-range luma and color values by eye. You can read pixel RGB values directly off your Window dialogs, VirtualDub, and other viewers (but not media players) with a free (!) pixel reader like Csamp (http://www.netreach.net/~gavin/gavsfreeware/csamp.htm ).

I've always wondered why Avery Lee doesn't add the feature directly to VirtualDub. It's such an obvious necessity, and simple to implement. The RGB (and YUV) values could be displayed along with the timeline information.

[jmac698]

AvsPmod has this feature, which I requested actually

[themaster1]

Does it ? which version i have missed that feature if true

[Gavino]

Does it ? which version i have missed that feature if true

It's been there at least since an earlier version of the original AvsP.
Options->Program Settings->Video 1->Customize video status bar.

See also Help->Video Features->Customizing status bar message.

EDIT: checked the history in the readme file and it was introduced in v1.3.7 of AvsP (March 2007).

[jmac698]

Yes and I requested it here http://forum.doom9.org/showthread.php?p=899174
But note, there is a bug in the latest version that it's reading the wrong coordinates if the preview is zoomed.

[jwbrasil2]

You have to check some VB filters able to fine tune color and brightness like old color restoration, hue, hsv, smart brightness, color equalizer...

[sanlyn]

You have to check some VB filters able to fine tune color and brightness like old color restoration, hue, hsv, smart brightness, color equalizer...

Maybe I missed something, but I'm not sure what this image is supposed to demonstrate. It's badly discolored and the levels are shot.


[Attachment 9721 - Click to enlarge]

[Abbadon]

This hobby is kind of tough to learn, I am still trying, but things don't look very promising, this is the latest raw capture from VHS with the before and after:

[sanlyn]

This hobby is kind of tough to learn, I am still trying, but things don't look very promising, this is the latest raw capture from VHS with the before and after:

Yes, it can get a little tricky. But there's, progress.

The "before" image (below) shows that you need to learn how to use a histogram and to read pixel values. Unless you have an exceptionally experienced eye , a couple of simple tools would have prevented some mistakes. The histogram for the "before" image could pojnt you in the right direction, even if some of the problems can't be solved in RGB (it's best to fix some of them in YUV first).


[Attachment 9727 - Click to enlarge]

The :before" image is losing detail and has a glaring look because, among other things, your luma range is crushed on one end and blown out at the other. Blue is stretched beyond spec and lost data at both ends, and red is starting to climb up the right-hand wall. There's one color that the picture really needs, but it isn't there: black. The cat's head isn't black; it's mostly purple. The tail is brown (is it supposed to be?). Discolored blacks are responsible for some of the softening or fading effect in the drawn lines, which should be closer to a clean black but aren't.

In the "after" shot, the histogram reveals that the problems in the "before" shot were made worse"


[Attachment 9728 - Click to enlarge]

All three colors are now blown away and losing data at both ends, and the "black" lines are starting to lose detail and solidity, especially in bright areas. You should learn to control your levels and chroma spread, and to use histograms and pixel samplers to help you with that task.

http://trevlac.us/colorCorrection/colorTools.html
http://www.netreach.net/~gavin/gavsfreeware/csamp.htm
http://www.sphoto.com/techinfo/histograms/histograms.htm
http://www.sphoto.com/techinfo/histograms/histograms2.htm

[sanlyn]

You should be able to use an AviSynth script:

AviSource("filename.avi")
ConvertToYUY2(interlaced=true)
ColorYUV(cont_y=-36, off_y=-2)

You can come close to the same result in VirtualDub alone but you have to be careful. If you apply the Brightness/Contrast filter first it will usually work because that filter can work with the video in YUV. If you apply some other filter before it, the video will usually be converted to RGB. That is where the crushing of the Y<16 and Y>235 happens.
. . . . .


[Attachment 9661 - Click to enlarge]

Question: if 'y' is already too low, why use off_y=-2? Wouldn't that bring the bottom luma level farther down?

What to do about the out-of-spec chroma? The histogram says detail is being lost at the extremes. Shouldn't there be a gradual falloff at the extremes instead of butting against both edges of the graph? I've tried taming the extremes with gradation curves using its YUV mode. But I suspect the curve converts to RGB first, then to YUV, then back to RGB. That would be one way to keep color data at the low and bright ends (?). If you try to tame the extremes of Red and Blue in YUV, you end up with distorted Green - unless I'm just doing it wrong with the u and v commands.

[poisondeathray]

Question: if 'y' is already too low, why use off_y=-2? Wouldn't that bring the bottom luma level farther down?

By itself , yes, but he's reducing contrast as well

Contrast reduction reduces the high & low point of the waveform, usually from a center point, but some filters do it differently

What to do about the out-of-spec chroma? The histogram says detail is being lost at the extremes. Shouldn't there be a gradual falloff at the extremes instead of butting against both edges of the graph? I've tried taming the extremes with gradation curves using its YUV mode. But I suspect the curve converts to RGB first, then to YUV, then back to RGB. That would be one way to keep color data at the low and bright ends (?). If you try to tame the extremes of Red and Blue in YUV, you end up with distorted Green - unless I'm just doing it wrong with the u and v commands.

Yes, I don't think vdub's YUV curves works in YUV directly.

You can salvage quite a bit by either using full range for the Y'CbCr => RGB conversion matrix with avisynth then adjust in RGB, or adjust Cb and Cr in Y'CbCr space before bringing it into vdub

Because you have no control over the colorspace conversion in vdub (it's using Rec601), the chroma can get clipped, because the matrix specfices

CbCr [16,240] => RGB [0,255] , so the CbCr values <16 and > 240 are lost

Note, even a full range matrix will not necessarily recover all the information in the Y'CbCr source, because not all values have a valid representation in sRGB . There are wider gamut RGB spaces, but not in vdub.

[jagabo]

ColorYUV(cont_y=-36, off_y=-2)

Question: if 'y' is already too low, why use off_y=-2? Wouldn't that bring the bottom luma level farther down?

cont_y=-36 reduces the contrast (darks get lighter, lights get darker) so it pulls the black level up. off_y=-2 is used to pull it down a bit.

What to do about the out-of-spec chroma? The histogram says detail is being lost at the extremes. Shouldn't there be a gradual falloff at the extremes instead of butting against both edges of the graph?

If you're talking about the peaks in the RGB histograms in that same post:

https://forum.videohelp.com/threads/340808-Capturing-from-old-VHS-and-improving-quality...=1#post2121186

Those are from the blown out whites. They have nothing to do with the chroma. Remember, white = R,G,B 255,255,255. So any image with a lot of white will have peaks at the high end of the RGB histograms. U and V graphs of the same video:



But I wasn't even looking at colors in the hopes that he could capture again without the blown out brights and crushed blacks.

[Gavino]

if 'y' is already too low, why use off_y=-2? Wouldn't that bring the bottom luma level farther down?

To put it in numerical terms, the input and output luma values are related as follows (assuming unchanged gamma for simplicity):
out = in + in*gain_y/256 + (in-128)*cont_y/256 + off_y
which produces the result described by poisondeathray and jagabo:

By itself , yes, but he's reducing contrast as well.
Contrast reduction reduces the high & low point of the waveform, usually from a center point, but some filters do it differently

cont_y=-36 reduces the contrast (darks get lighter, lights get darker) so it pulls the black level up. off_y=-2 is used to pull it down a bit.

[sanlyn]

OK, got it. The blown-out colors are why I'm going to recapture a VHS again with different proc amp settings than the first time, just for the usual scenes you get on VHS with far out color loss. The "normal" scenes will look crummy, but I'll build the VHS transfer using the best of each capture. Not that I haven't had to do this before(!).

Okay, makes more sense now. Appreciated.

[jagabo]

I don't know what the colors are supposed to be in that video clip. But I suspect this is in the ballpark:

AviSource("Dingbat_land_sample.avi")
ColorYUV(cont_y=-36, cont_u=100, cont_v=200)
ConvertToYV12()
McTemporalDenoise(settings="very high")

Note, I just treated the video as progressive for the purposes of this color sample.

[sanlyn]

[QUOTE=jagabo;2122344]I don't know what the colors are supposed to be in that video clip. But I suspect this is in the ballpark:

AviSource("Dingbat_land_sample.avi")
ColorYUV(cont_y=-36, cont_u=100, cont_v=200)
ConvertToYV12()
McTemporalDenoise(settings="very high")

Yes, they do look about right. I'm looking at your contrast settings, though. So it's allowed to push chroma far beyond 16-235, as long as luma stays inside that range? No wonder I can't get good color out of AviSynth, I've been doing the opposite with colors.

[jagabo]

So it's allowed to push chroma far beyond 16-235, as long as luma stays inside that range?

There were no illegal chroma values after processing. I don't have the video any more but here's the U and V traces of the JPG image:



The original video had very pale colors. I just increased the saturation (chroma contrast, cont_u, cont_v) and decreased the luma contrast (cont_y). Legal values for chroma range from 16 to 240.