mpg2 color correction?

[jagabo]

Some other thoughts for the OP:

Setting the peaks at IRE 100 (in post 26) was wrong because the true peaks in the video were hacked off by the over bright capture (either on the laserdisc or by the DVD recorder). I should have shot lower.

When working in YUV, reducing the brightness (Y) results in a natural increase in saturation. Because the chroma channels are essentially subtracted from the brightness. For example, if you have the equivalent to RGB 250,250,250 and your U and V values result in 50 being subtracted from the red and blue you're left with 200,250,200, a not so saturated green. But if you start with 100,100,100 and subtract the same 50 units from red and blue (ie, the same U and V values) you get 50,100,50, a much more saturated (and darker, obviously) green.

[Mike99]

Some other thoughts for the OP:

Setting the peaks at IRE 100 (in post 26) was wrong because the true peaks in the video were hacked off by the over bright capture (either on the laserdisc or by the DVD recorder). I should have shot lower.

When working in YUV, reducing the brightness (Y) results in a natural increase in saturation. Because the chroma channels are essentially subtracted from the brightness. For example, if you have the equivalent to RGB 250,250,250 and your U and V values result in 50 being subtracted from the red and blue you're left with 200,250,200, a not so saturated green. But if you start with 100,100,100 and subtract the same 50 units from red and blue (ie, the same U and V values) you get 50,100,50, a much more saturated (and darker, obviously) green.

I have been trying the various scripts that you have provided.

When watching a longer clip with adjusted/enhanced colors there appears to be a very slight “pumping” or varying of the colors. I also notice this when viewing the VHS sourced video. I do not notice this when watching the original LD/DVD version, perhaps because the colors are weak. It may be happening but is not apparent. I’ll try to dig out & connect the laser disc player this weekend & watch the original LD.

In the original LD/DVD some scenes look OK & some look faded. Therefore I was trying different values in your scripts looking for a possible one-size-fits-all compromise. I’m learning a few things about the “ColorYUV” script in the process, but am having difficulty reducing what I think is a slight greenish cast in some scenes. I’m doing a lot of trial & error using different values for U and V and Y, but feel like I’m chasing myself in circles. I increase the U and V, but then the image looks too dark of contrasty. And then try to reduce Y in order to compensate. But I’m not sure if I should be changing gain or contrast or saturation or offset (which I’m not exactly sure how that functions). Any suggestions how to slightly reduce green?

Thanks in advance.

[jagabo]

When watching a longer clip with adjusted/enhanced colors there appears to be a very slight “pumping” or varying of the colors. I also notice this when viewing the VHS sourced video. I do not notice this when watching the original LD/DVD version, perhaps because the colors are weak. It may be happening but is not apparent.

Yes, I saw it in the short clip too. The colors are shifting as the scene is changing. I think brightness of the picture is causing the white balance to change.

In the original LD/DVD some scenes look OK & some look faded. Therefore I was trying different values in your scripts looking for a possible one-size-fits-all compromise. I’m learning a few things about the “ColorYUV” script in the process, but am having difficulty reducing what I think is a slight greenish cast in some scenes. I’m doing a lot of trial & error using different values for U and V and Y, but feel like I’m chasing myself in circles. I increase the U and V, but then the image looks too dark of contrasty. And then try to reduce Y in order to compensate. But I’m not sure if I should be changing gain or contrast or saturation or offset (which I’m not exactly sure how that functions). Any suggestions how to slightly reduce green?

Move U and V up with off_u and off_v. Pure white/grey has both U and V at 128.

[jagabo]

There's one other thing you might try: copying the VHS chroma channels onto the laserdisc luma. The colors won't be as sharp. You'll have to adjust the frame sizes/cropping to get the channels to line up. And you'll have to make sure the frames match. Here's an example using the two images in post #21, luma from the laserdisc (levels modified) and chroma from the VHS:



I didn't resize and position the chroma exactly, but you get the idea.

v1=ImageSource("Laser.png").ConvertToYUY2()
v2=ImageSource("VHS.png").BilinearResize(720,540). ConvertToYUY2()

v1=ColorYUV(v1, gain_y=-70, off_y=15) #adjust levels a bit
v2=ChromaShift(v2,c=-10, l=4) #shift the chroma left and down

MergeChroma(v1,v2)

[themaster1]

Work with the vhs cut, i propose this:




Or this perhaps:

[sanlyn]

Sounds like a lot of mind-bending work. IMHO working with the original YUV is the best way (sometimes the only way) to initially bring color and levels into line with "normal" standards -- especially considering the way so-called "pro" commercial disc/tape is often sloppily thrown together. But tweaking specific color areas, etc., -- most upscale apps (Premiere, AE, etc. and their expensive plugins) use RGB. But then there are shops that write their own YUV-specific code to handle particular problems. I don't know that most people would have time or wherewithal to go there.

[Mike99]

jagabo -

I don’t know if the frames can be made to match because the PAL VHS was converted to NTSC DVD and the playing time is shorter than the NTSC LD. I’m presuming the 25fps PAL was sped up to 24fps which is what I read usually happens.

The NTSC LD plays 94mins, 20secs.
The PAL>NTSC plays 90mins. 30secs.

sanlyn -

I’ve never adjusted YUV color until now, so this is all new to me. From my previous dabbling with video I’ve read that YCbCr is digital component video. And that YPbPr is analog component video which is what most consumer electronics use.

I still want to view my LD on the HDTV to see how the colors look but did not have a chance to drag out the LD player this past weekend (I had to work on the car). But I’ll try to do this shortly.

[Gavino]

Now, if someone can work up an AViSynth plugin that's as easy to use as the RGB "Levels" histograms in VirtualDUb/Adobe/Vegas, it would make AviSynth so much easier to play with. Yes, I know, you have ColorYuv (gamma_y, luma-this, contrast-that, etc.) and Tweak and whatnot. A hassle from the DOS error; workable, but fiddly and primitive. Levels histograms use sliders, some of them have sliders for 4 channels (R,G,B, and RGB master). Slide here, slide there, nudge a bit to tweak, you're done. Fixed in 2 minutes, compared to playing numbers soccerball in AviSYnth for half an hour.

If you use AvsP (or AvsPMod) to edit your Avisynth scripts, you get sliders.
Have you given up on that for some reason?

[jagabo]

IMHO working with the original YUV is the best way (sometimes the only way) to initially bring color and levels into line with "normal" standards -- especially considering the way so-called "pro" commercial disc/tape is often sloppily thrown together.

This is because analog video is in YUV so the errors analog equipment makes are YUV errors. (Yes, I'm loosely equating YUV, YCbCr, YIQ, etc.)

The most interesting aspects of all this reading and looking is that no one seems to agree on how accurately the transitional filters and software can work with YUV->RGB>YUV processing. Even some $10,000 gear seems to make conversion errors that require human intervention

There are several reasons for this.

1) The input and outputs are usually 8 bits (per channel). That means you have rounding errors. Conceptually, with highly simplified math:

1 / 3 = 0.333...

But we are working in integers so you can't have 0.333... it gets round to 0. Then when we reverse the operation we start with that 0 and:

0 * 3 = 0

So what was originally 1 is now 0. If you repeat the conversions starting with 0 you'll see that the result is also 0. So not every RGB value maps to a unique YUV value, and vice versa.

2) The usual rec.601 or 709 matrix uses a limited range of values. 16-235 for Y, 16-240 for U and V. That means there are only 11,376,500 different colors in YUV vs. RGB's 16,777,216.

3) Not every possible 8 bit RGB value maps to the 8 bit YUV colorspace, and vice versa. Mathematically it's the equivalent of:

255 * 1.2 = 306

But the converted value must be within the range of 0-255. So 306 becomes 255. But the same thing happens to 254, 253, 252, etc.

254 * 1.2 = 304.8
253 * 1.2 = 303.6
252 * 1.2 = 302.4

All of those also get clamped to 255. Ie, the bright, saturated corners of the RGB cube get cropped when converted to YUV.

Even good ole NeatVideo poses an odd problem: it accepts and outputs RGB only, but internally it works in its own proprietary YCbCr (which, I find, is not the same as YUV, which YCbCr is often referred to in error).

Yes, YUV and YCbCr are different (the weighting of the color coefficients is different) . But they are similar in that they are subtractive color systems. Ie, you have a grayscale image (Y) and you produce color by subtracting the chroma channels (U and V, or Cb and Cr) from it. Whereas in RGB systems you start with black and add R, G, and B values to produce color. The reason analog video uses YUV instead of RGB was to maintain compatibility with black and white TV broadcast. The Y component is the same as the black and white signal for black and white TVs. Colors TVs mix in the U and V components to make color. So the same signal can be displayed by both.

So the best I come up with is to accomplish whatever can be done in YUV. If RGB is necessary...

On the other hand we have film which works in RGB. There are tiny red, green, and blue crystals in the film that make colors. Or three rolls of black and white negatives (one for red, one for green, one for blue). They may fade at different rates. These types of problems are usually best solved in the RGB color space because the errors are in RGB. Say for example the blue channel has faded by 37 percent and the red channel has faded by 17 percent. It's easy to boost the blues and reds individually to restore them to their original values. But in YUV red, green, and blue are not completely independent. It's not possible, or at least very difficult, to change one without changing the others.

Meanwhile, what I see of the LD version of this scene we're all playing with has so many problems, it makes you question which is worse, LD or VHS.

Which is why I suggested he go back and look at the laserdisc to see if the problems were on the disc or caused by his DVD recorder. If some of the problems were caused by the DVD recorder maybe he can find an alternate capture method.

[jagabo]

I don’t know if the frames can be made to match because the PAL VHS was converted to NTSC DVD and the playing time is shorter than the NTSC LD. I’m presuming the 25fps PAL was sped up to 24fps which is what I read usually happens.

If that was the only issue it could easily be solved. But it's not likely to be the only problem.

[Gavino]

Not every possible 8 bit RGB value maps to the 8 bit YUV colorspace, and vice versa.
...
the bright, saturated corners of the RGB cube get cropped when converted to YUV.

That's wrong.
Every possible RGB value does map to a valid YUV value (even the "bright saturated corners").
This is independent of whether you are mapping to 16-235 or the full range 0-255.
It's only the YUV->RGB mapping that can produce out-of-range values (resulting in one or more of the RGB components being clamped at 0 or 255).

In fact, less than half of the YUV space corresponds to valid RGB values, especially for low and high values of Y, and for no value of Y is the entire range of U or V valid. For the extreme values of Y (corresponding to black or white), the only possible values for U and V are U=V=128.

[jagabo]

Every possible RGB value does map to a valid YUV value (even the "bright saturated corners").

Yes, it looks like I remembered incorrectly. Here's are some pictures that show the RGB cube fully inside the YUV and YCbCr color cubes, and the equations used to convert them:
http://software.intel.com/sites/products/documentation/hpc/ipp/ippi/ippi_ch6/ch6_color_models.html

[sanlyn]

[QUOTE=jagabo;2114080]
This is because analog video is in YUV so the errors analog equipment makes are YUV errors. (Yes, I'm loosely equating YUV, YCbCr, YIQ, etc.)

1) The input and outputs are usually 8 bits (per channel). That means you have rounding errors. Conceptually, with highly simplified math: ...

I quoted just a little here, jagabo, just to say hat this is the same info (and quandaries) I'm finding. Shucks. I guess you just have to work out these captures and restorals piece by piece. But it does help to get a handle on where the problems come from. Now if I just had $150,00 worth of studio gear . . . Sorry for the digression, but thanks for clarifying.

[Gavino]

... the YUV->RGB mapping that can produce out-of-range values (resulting in one or more of the RGB components being clamped at 0 or 255).

In fact, less than half of the YUV space corresponds to valid RGB values, especially for low and high values of Y, and for no value of Y is the entire range of U or V valid. For the extreme values of Y (corresponding to black or white), the only possible values for U and V are U=V=128.

Some time ago, I constructed a graphic demonstration of this using Avisynth:
U and V ranges for valid RGB

In that same thread (post #14), I also wrote a function to detect and display the 'invalid' YUV pixels in a video (those that map to out-of-range RGB). This could be useful after a session of tweaking in the YUV domain to check that you haven't gone outside the range of valid RGB values.

[Mike99]

I never thought about the movie’s colors until a few weeks ago when someone mentioned there was very good Canadian DVD version, but they didn’t know the source material. I have the Japanese LD, and there never was a NTSC DVD version. So I thought I’d look at my LD>DVD & compare it to the PAL VHS to DVD version someone gave me. Then the project began.

I did some more Googling and found this:
http://www.disneymovieslist.com/movies/song-south.asp
If you click on the trailer video it sort of looks like my LD/DVD, including those little flecks of noise or dirt.

Then I came across this. It looks like someone has spent a lot of time trying to make a very good copy of this movie.
http://atlas.kennesaw.edu/~dhirschl/song_of_the_south/

Looking at the above it appears the Japanese LD is poor to begin with. I was happy to own it & never questioned the quality because it was the only version available that I knew about many years ago. I bought it new from a LaserDisc store. Remember those?

It looks like correcting all the LD defects is quite a task, but I’d still like to make a few tweaks to improve its colors.

[jagabo]

Some time ago, I constructed a graphic demonstration of this using Avisynth:
U and V ranges for valid RGB

In that same thread (post #14), I also wrote a function to detect and display the 'invalid' YUV pixels in a video (those that map to out-of-range RGB). This could be useful after a session of tweaking in the YUV domain to check that you haven't gone outside the range of valid RGB values.

Nice work. I added the functions to my library.

[themaster1]

Which filters are required for your Valid RGB script gavino, i get a no function mt_lutspa error. I have loaded masktools-v2.0a36 so i don't know what's wrong

[intracube]

I was able to recover some bright details. And I adjusted the colors a little different than last time:

[Attachment 9130 - Click to enlarge]

Nice work there jagabo

When you were looking at the full YUV range, did you see any evidence of the image clipping at some level above Y=235, or did it go all the way to 255?

The reason I ask is video on laserdiscs is analogue, right? Is it possible that the video on the disc is out of spec and overloading the LD player. Or maybe the player itself is outputting a signal outside the legal range which is overloading his capture device.

Little can be done if it's the former (short of trying a different LD player), but if it's the latter, Mike could try a basic video processing box inline between the LD player and whatever he's using to digitise the video to bring down the levels.

Mike; What did you use to digitize the LD video in the first place? A DVD recorder?

Apologies if any of this has been said already - I've only skimmed this thread.

[jagabo]

When you were looking at the full YUV range, did you see any evidence of the image clipping at some level above Y=235, or did it go all the way to 255?

The luma values went all the way up to 255. It was obvious the peaks would have been over 255 but they were clamped, of course.

The reason I ask is video on laserdiscs is analogue, right? Is it possible that the video on the disc is out of spec and overloading the LD player. Or maybe the player itself is outputting a signal outside the legal range which is overloading his capture device.

Yes, it's possible the problem is on the laserdisc. The DVD recorder (or capture card) would look at the depth of the horizontal sync pulse to adjust the automatic gain. If the pulse wasn't low enough there would be too much gain applied to the picture.

Little can be done if it's the former (short of trying a different LD player), but if it's the latter, Mike could try a basic video processing box inline between the LD player and whatever he's using to digitise the video to bring down the levels.

Yes, an analog proc amp might help.

[Gavino]

Which filters are required for your Valid RGB script gavino, i get a no function mt_lutspa error. I have loaded masktools-v2.0a36 so i don't know what's wrong

Masktools is the only dependency, and it should work with v2.0a35 or above, so it's odd that you are having problems..

Make sure you are using only the right dll for your Avisynth version:
mt_masktools-25.dll for Avisynth 2.5x
mt_masktools-26.dll for Avisynth 2.60
and that you do not have both of these together in your plugins folder.

[themaster1]

I have downloaded the mt 26 for avisynth 2.6 but i still get an error "not function mt_lutspa"

[Gavino]

I have downloaded the mt 26 for avisynth 2.6 but i still get an error "not function mt_lutspa"

Strange.
Are you sure it's an up-to-date version (35 or later)?
Are you loading it explicitly via LoadPlugin, or using the automatic load from the plugins folder?
Both should work, so I've no idea why it isn't.

Can you call other masktools functions?

[Mike99]

I was able to recover some bright details. And I adjusted the colors a little different than last time:

[Attachment 9130 - Click to enlarge]

Nice work there jagabo

When you were looking at the full YUV range, did you see any evidence of the image clipping at some level above Y=235, or did it go all the way to 255?

The reason I ask is video on laserdiscs is analogue, right? Is it possible that the video on the disc is out of spec and overloading the LD player. Or maybe the player itself is outputting a signal outside the legal range which is overloading his capture device.

Little can be done if it's the former (short of trying a different LD player), but if it's the latter, Mike could try a basic video processing box inline between the LD player and whatever he's using to digitise the video to bring down the levels.

Mike; What did you use to digitize the LD video in the first place? A DVD recorder?

Apologies if any of this has been said already - I've only skimmed this thread.

Yes, I did a direct LD player to DVD recorder copy.

Instead of using a basic video processing box, is there a way to do this if capturing to a PC? IOW can a video capture card make analog adjustments before the signal is digitized? Or is it already too late once the card has captured the video?

[jagabo]

can a video capture card make analog adjustments before the signal is digitized?

Yes, some can. Actually, they may make the adjustments in digital form but they may capture a wider range with with more than 8 bit precision so that may get around the problem. It depends on how far out of spec the signal is. There's one other possibility, a macrovision stripper, or "video clarifier". They reconstruct the horizontal sync pulse in order to strip macrovision. That might fix the problem. DVD recorders are designed to respond to macrovision in various ways. An overbright picture and bad colors are some of them.

[sanlyn]

Now, if someone can work up an AViSynth plugin that's as easy to use as the RGB "Levels" histograms in VirtualDUb/Adobe/Vegas, it would make AviSynth so much easier to play with. Yes, I know, you have ColorYuv (gamma_y, luma-this, contrast-that, etc.) and Tweak and whatnot. A hassle from the DOS error; workable, but fiddly and primitive. Levels histograms use sliders, some of them have sliders for 4 channels (R,G,B, and RGB master). Slide here, slide there, nudge a bit to tweak, you're done. Fixed in 2 minutes, compared to playing numbers soccerball in AviSYnth for half an hour.

If you use AvsP (or AvsPMod) to edit your Avisynth scripts, you get sliders.
Have you given up on that for some reason?

True, you can make your own sliders. But they don't "work" like VDub or Adobe or Vegas 3-point level histogram controls. Adobe's versions are the more sophisticated, with 3-point sliders for each RGB channel + master. I realize those controls are the equivalent of separate brightness, gamma (midtone) and contrast controls but they're interactive and hooked to the histogram(s).

BTW I am using AvsPMod for many tasks.

[Mike99]

I watched the LD directly on the HDTV and it looked better than watching the DVD copy on the PC. But the DVD copy on HDTV also looked better than watching it on the PC. I really could not tell any difference between the LD & the DVD copy.

I also watched the VHS > DVD on the HDTV. I thought the colors were over saturated in most of the animated scenes and some of the real people scenes. And the color “pumping” was more noticeable than the LD or LD>DVD versions.

Overall I’d have to say I like the LD version better. Again it’s not perfect, but the scenes with weak color are outnumbered by the VHS scenes with over saturated color.

I don’t know why the LD/DVD colors look so much lighter on my PC’s monitor. I don’t expect them to look exactly the same, but the difference is quite a bit. Both have been adjusted using various test patterns so they should not be too far off. Maybe it’s just the inexpensive PC monitor.

Anyway the short clip scene that I posted is perhaps the weakest as far as color. I’ll still play around & see how much I can tweak it & similar scenes without over saturating others.

Thanks to everyone for all the input & suggestions.

[sanlyn]

Try VLC Player.

Didn't you say the LD was Japanese? I believe they use different IRE levels than North America, don't they? Again, folks, correct me on that.

[jagabo]

Didn't you say the LD was Japanese? I believe they use different IRE levels than North America, don't they?

Yes, the North American analog NTSC signal includes 7.5 IRE "setup" and Japan uses 0.0 IRE setup. Ie, the black level is lower in Japanese signals. Watching a Japanese NTSC analog signal on an American TV would result in too much contrast with crushed blacks.

But I don't know what happens with lasediscs. Is the content of the NTSC-US and NTSC-J discs the same and it's the players job to add the setup? That's the case with DVDs where the data is digital. But the signal recorded on a laserdisc is analog composite. And laserdisc players were pretty dumb machines.

Mike99's laserdisc sample shows contrast problems with both the black and white levels. DVD recorders often have compensation for the difference between NTSC-J and NTSC-US. It's usually called "dark" and "light" or something like that.

[Mike99]

I guess I never mentioned it was a Japanese LD. Sorry about that. With most things being HD digital nowadays it’s been a long time since I’ve thought about the different IRE levels.

I just did some Googling & it appears most Japanese LDs are 0 IRE. One person determined one of his was 7.5 IRE, but that appears to be the exception from what I found so far.

I have a Panasonic DVD recorder which has an input level setting which “Adjusts the black level of the input signal“. There is a choice of either Lighter or Darker, and Darker is the default. I’m going to change this to Lighter and copy some of the LD again and see what it looks like.

Regarding the PAL VHS to DVD version: Some of it’s colors look over saturated when watching on the HDTV. If this PAL source was 0 IRE and transferred as if it was 7.5 IRE, could this account for the over saturated look on the HDTV?

PS - I was using VLC player when watching on the PC.

[Mike99]

My Panasonic DVD recorder has input level settings of either Lighter or Darker, and Darker is the default. I fed the video through the Panny & changed the input setting to Lighter. This does lighten up the dark & shadow areas. If black crush were eliminated there should be more shadow detail, but I cannot perceive any increase in detail. And the picture looks a bit washed out compared to the Darker setting. But I don't know which is more accurate.

At the end of the LD there are a few clips advertising other movies & some of these are letterboxed. With the Darker setting the black letterbox frame is almost the same as the black pillars on either side. With the Lighter setting the letterbox frame is dark gray and does not match the “true” black of the pillars.

So I’m not sure which setting is correct.

Is there a way to determine which black level is used on the LD?