Have we had DVD colors wrong this whole time?

[flossy_cake]

I used to have episodes recorded off broadcast TV but I deleted them (doh) , would be very useful to compare the colour. But the TV network was using old PAL analogue masters from that era so the colour conversion to PAL was probably bodged anyway

[flossy_cake]

https://i.imgur.com/atnnhtjl.png
https://i.imgur.com/ZR72eBTl.png


The laserdisc version above looks good here to me - perhaps that should be the reference?

The DVD version has this gross dimness and oliveness to it. It's clear to me something went wrong with the master used for the DVDs.

[wonkey_monkey]

Bear in mind the Laserdisc "has no color standard flags so After Effects interprets it as Rec.709", which it shouldn't do.

Again it all comes down to personal preference really.

[flossy_cake]

Bear in mind the Laserdisc "has no color standard flags so After Effects interprets it as Rec.709", which it shouldn't do.

Again it all comes to personal preference really.

Yeah the difference between 709 and 170M is not that much on a typical scene especially with these low luminance colours, previously I described it as a "just noticeable difference". In madVR it's easy to toggle between them with a hotkey

Anyway I'm looking at those 2 screenshots on my calibrated S90D and on this TV the difference between DVD and laserdisc isn't as great as it is on my IPS PC monitor. Both are calibrated with a meter but the S90D is more extensively calibrated in terms of colour space and saturation tracking linearity

I'll probably just go with +10 saturation clicks in madVR for season 1 of DS9, seems to look okay-ish. Maybe +15 for certain episodes.

[flossy_cake]

Here are some comparisons from DS9 S01E09 The Passenger. It's the standard 170M/SMPTE-C gamut and the correction is in MadVR's controls: saturation increased by 15 clicks, hue rotated towards red by 4 clicks. I am trying to get rid of this pale olive tone in actors faces. In hindsight perhaps saturation 12 and hue 3 might be preferable.

Best to view these full screen in a dim surround and quickly swipe the bar to get the instant swap effect

https://imgsli.com/Mzk5MDMw
https://imgsli.com/Mzk5MDMx
https://imgsli.com/Mzk5MDMz
https://imgsli.com/Mzk5MDM0
https://imgsli.com/Mzk5MDM2
https://imgsli.com/Mzk5MDM3
https://imgsli.com/Mzk5MDM4
https://imgsli.com/Mzk5MDM5

[Sunspark]

DS9 came out on retail VHS tapes, and obviously those will not be high resolution, but it WOULD be interesting to compare the colours off a VHS tape.

So, digging into this, since we have already been talking about master tapes in production.

VHS used YIQ. Y for luma, I and Q for chrominance. I was for orange-blue hue axis and Q was for green-purple hue axis.

YIQ used the NTSC 1953 RGB primaries which is not the same as rec.601 or sRGB.

https://en.wikipedia.org/wiki/YIQ

Then there's NTSC itself, different white point, different primaries.

https://en.wikipedia.org/wiki/NTSC#Colorimetry

Says that SMTPE-C wasn't until 1994 and that places like Japan stayed on the 1953 one.

"To adjust for proper viewing requires gamut mapping via LUTs or additional color grading. SMPTE Recommended Practice RP 167-1995 refers to such an automatic correction as an "NTSC corrective display matrix."[38] For instance, material prepared for 1953 NTSC may look desaturated when displayed on SMPTE C or ATSC/BT.709 displays, and may also exhibit noticeable hue shifts. On the other hand, SMPTE C materials may appear slightly more saturated on BT.709/sRGB displays, or significantly more saturated on P3 displays, if the appropriate gamut mapping is not performed."

I feel the images are desaturated and hue-shifted so 1953 NTSC needs to be explored.

[flossy_cake]

I feel the images are desaturated and hue-shifted so 1953 NTSC needs to be explored.

1953 is 470M, and decoding with 470M has been extensively explored, and on many scenes it looks subjectively better, but on some other scenes it looks obviously wrong. That's why I don't think the intended gamut is actually 470M.

Another observation is that the later seasons of DS9 don't seem to suffer from major issues and look ok at the standard 170M/SMPTE-C gamut, it's only the early seasons that have this paleness and green hue.

This got me thinking about the hue control. TVs still have a hue control as a carryover from analogue days, apparently this is to allow the user to compensate for an inherent issue with analog NTSC signals where phase errors in the signal can cause hue shifts and this was most noticeable on skin tones being too green or too magenta. Apparently PAL doesn't suffer from this issue and that was touted as a key advantage of PAL. So I think this NTSC hue issue could be something to investigate deeper.

Earlier I wrote that adjusting hue didn't give me a subjectively pleasing result, but I think that's because I was trying to adjust it too much. Only 3 or 4 clicks in MadVR seems to be the limit, any more and then certain other scenes started to look wrong.

So for the early seasons of DS9 try MadVR saturation +12 to +15 , and hue 3 to 4 clicks red. I would be interested to hear some more opinions on this.

[flossy_cake]

Also there is boosting of saturation, and then there is boosting of chroma.. Analogue signals use chroma not saturation, so I will try some avisynth filters to boost chroma instead of saturation to see how that looks instead of using madVRs saturation control

[Sunspark]

Now that I think about it, I do remember adjusting the hue and sat on my old CRT, though I couldn't tell you now what it would have been.

Chroma is interesting, especially for reds. The algorithm chosen makes a difference as it can make some colours hotter/more saturated.

[flossy_cake]

@DS9Redefined

I had a random memory where I was directly playing a DVD disc from season 1 of the region 1 version of Dr Quinn Medicine Woman and I had the Ctrl+J debug screen up in MPC-HC and I seem to recall being surprised the colours appeared more saturated and the gamut was being reported as 470M. This gamut metadata was only available when playing the disc directly from the DVD drive in MPC-HC using the built-in DVD Navigator filter. If I remuxed the mpeg2 video stream from the disc into mkv files using MakeMKV software, that gamut metadata was lost and it just defaulted to 170M/SMPTE-C. If I opened the VOB file directly, same thing.

So, if you have the original disc of S01E01E02 The Emissary, could you try opening the disc itself in MPC-HC with DVD Navigator as the source filter and LAV as the video decoder, hit ctrl+J to bring up the debug screen and check this metadata highlighted in yellow:


https://i.imgur.com/JlhUCo2.png


Note: in the above image it says 170M cause my season 1 discs are on loan to a family member, but I should get those back in a few days.

btw there is ffprobe.exe as an alternative way of getting this gamut metadata directly from the disc. It is included alongside ffmpeg.exe but I couldn't get it to work

[flossy_cake]

On this scene for example, The Emissary @ 33:20 , the 470M decoding appears to be the correct one. The right image is 170M with the aforementioned hue and saturation corrections, but I think it still looks quite wrong compared to 470M without any such corrections (left image)

https://imgsli.com/Mzk5NTkw

Maybe the scene to scene variance comes from NTSC = Never The Same Colors?

[wonkey_monkey]

The right image is 170M with the aforementioned hue and saturation corrections

How does it look without?

[flossy_cake]

How does it look without?

Even more pale and greenish. My corrections were hue towards red 4 clicks and saturation increase 15 clicks, using the MadVR controls.

I just completed watching S01 E01E02 The Emissary in its entirety at 470M and scrutinising every scene, and I am now 100% convinced 470M is actually the intended gamut for that pilot episode.

I don't know if all episodes of season 1 were mastered at 470M - we need someone with the original region 1 NTSC discs to look for metadata that could possible clear this up.

[flossy_cake]

There may also be some differences in colour due to differences in the way various renderers perform gamut conversion. Like, relative vs perceptual intent.

Plus, if using MadVR we must remember to go in the settings and choose "calibration > this display is already calibrated to the following gamut /primaries > 709" (replace 709 with whatever your display's gamut is, obviously, but it's typically 709). Otherwise MadVR won't perform any gamut conversions

[Sunspark]

My library has the disc but I'll wait to see if DS9Redefined has it already.

>If I remuxed the mpeg2 video stream from the disc into mkv files using MakeMKV software, that gamut metadata was lost and it just defaulted to 170M/SMPTE-C.

This is troubling. We don't want important flags to be lost.

[flossy_cake]

This is interesting...here's an example from S01E12 Vortex where it's the other way round: 470M seems to look wrong and 170M looks ok:
https://imgsli.com/Mzk5NjI2

Now compare it to the 470M decoding of S01E01E02:
https://imgsli.com/Mzk5NjI3

The skin tone is very similar, so I think 470M is correct for S01E01E02 and 170M is correct for S01E12

[flossy_cake]

It occured to me that if we have 2 sample episodes, 1 of each known gamut, we should then be able to use the opening titles as a reference to colour match them to any other episode to figure out its correct gamut. Particularly the hue and saturation of the gold lettering which is of particular use as it's within the same region as skin tones.

First, let's start by comparing Emissary at 470M vs Vortex at 470M. Under the premise that Emissary 470M is correct and Vortex 170M is correct, if we decode Vortex at 470M then we should find its gold lettering should appear excessively reddish and more saturated than Emissary 470M, and it does:
https://imgsli.com/Mzk5NjQy

Now let's check it the other way round and compare Emissary at 170M vs Vortex at 170M. Emissary 170M gold lettering should appear less reddish and less saturated, and it does:
https://imgsli.com/Mzk5NjQz

Now let's compare apples and apples: Emissary 470M vs Vortex 170M. These two should look identical, but they don't - Vortex 170M gold lettering is not quite as reddish and saturated as Emissary 470M:
https://imgsli.com/Mzk5NjQ0

At this point, if we say Emissary 470M is the correct decoding, we can use hue and saturation corrections to Vortex 170M to make it match the appearance of Emissary 470M. In this case, hue 4 clicks towards red, and saturation increased 15 clicks using MadVR's controls seems to match them:
https://imgsli.com/Mzk5NjQ1

[flossy_cake]

By the way, if you're using MadVR you can tag any file or folder name with [hue=-4] [saturation=15] [primaries=170M] for the 170M episodes, and just [primaries=470M] for the 470M episodes. To confirm primaries hit Ctrl+J, to confirm hue and saturation make hotkeys for them in MadVR settings and tap them once to show the current value.

[flossy_cake]

Ugh now I'm not so sure, because Sisko's uniform is way more pinkish magenta when using 170M with corrections so even though it matches the gold lettering on the title screen it's not matching Sisko's uniform. So that tells me something still isn't right.

edit: or is that Never The Same Colors at work?

[Sunspark]

Does it matter that the gamma seems different in the starfield?

[flossy_cake]

Does it matter that the gamma seems different in the starfield?

Probably, yes, but in what way I don't know. EOTF wasn't standardised until 2011 so the intended decoding gamma is a bit of a crap shoot for stuff made before then.

Another thing that's bothering me is if we look at Emissary 470M vs Vortex 470M title comparison, the blue glow underneath the runabout seems to match. This shouldn't be happening if Vortex's intended gamut is 170M cause 470M has a way different blue primary that is more teal.

Maybe instead I should try decoding Vortex as 470m and correcting it by reducing saturation (or chroma) and leave the hue alone.

[flossy_cake]

@DS9Redefined

Which episode of TNG is your first image in post#1 from? Because I'm looking at Picard's suit in the Bluray version and it's always RED never PURPLE, and this RED suit of his matches your 470M decode. So this means there are definitely episodes of TNG that are supposed to be decoded as 470M, which totally contradicts my comparisons on page 2. A google image search shows some shots where it appears PURPLE, others where it appears RED. But we know it's supposed to be RED because the Bluray is our "ground truth".

But then why does the skin tone seem so overly pushed on various scenes with 470M? I have no idea. Is it possible the colourist screwed up and boosted the chroma because they were incorrectly decoding with 170M on their preview monitor which produces dull greenish skin tones? This is just a random theory and probably not correct. Such a weird situation.

Anyway thanks for making this thread and to @Sunspark for notifying me of it - otherwise I probably would have done my inaugural screening of DS9 with wrong colours!

[flossy_cake]

I suspect the NTSC DVDs of TNG are suffering from that analogue NTSC hue phase shift making it look too magenta.
Here is my comparison again from page2 except this time with hue shifted towards green 4 clicks in MadVR

DVDrip 170M hue green +4
https://i.imgur.com/KmBAAlR.png

DVDrip 470M hue green +4
https://i.imgur.com/vcwhgVo.png

BDrip 709
https://i.imgur.com/hEIwjp2.jpeg

I actually think the 470M looks ok here with the hue correction.

Compare 470M hue correction on/off: https://imgsli.com/Mzk5ODg4

edit: it's from S01E05 12:00

[flossy_cake]

I've got to say, watching these Season 1 DVD episodes of TNG and DS9 in 470M transforms the show quite dramatically. DS9 for example went from feeling like some gritty dystopian run down space station with dull greys and browns to having all this lively colours that feels kind of magical. I'm starting to like it!

[Sunspark]

I enjoyed it back in the day. It had some excellent characters and episodes.

It's funny how watching old series set in the future years later there are things that are already anachronistic.. like for example an early episode where Jake Sisko asks if what he's looking at is (I forget) the station or Bajor or whatever. That Star Trek timeline didn't have it in mind that everyone might have their own personal AI with them.. there was only the ship's computer.

I do hope to see follow-up on the DVD Navigator filter thing.. because if the MakeMKV is not picking up the correct flag, that's a bug.

[flossy_cake]

https://imgsli.com/Mzk5OTEy

170M is puke!

(DS9 S01E04 7:42)

[huhn]

and the other is orange juice and obviously wrong so i do not follow.

[flossy_cake]

and the other is orange juice and obviously wrong so i do not follow.

That particular image of Dax does not look like orange juice to me on any of my displays. I can understand you saying the 470M image of Picard is oversaturated though. But the thing is you are comparing it to a less saturated image. I found that when actually watching the episode, after my eyes adjust it starts to look not so oversaturated. Vision is a very relative thing.

But even with 470M decoding on Season 1 of TNG I have seen some scenes where they cut to the next scene and all of a sudden Picards suit is purple instead of red, so there are mastering errors too.

Anyway, we need to see that flag on the disc with DVD Navigator filter :)

[huhn]

i have a calibrate display it's just orange juice in faces in all of the changed images.

and yes without the real frame this is really just even more guessing.
for best comparison compare the pal with ntsc like US, UK, FR, DE and IT. most of them should exist. pal has a better track record then NTSC or never twice the same colour.

[flossy_cake]

i have a calibrate display it's just orange juice in faces in all of the changed images.

and yes without the real frame this is really just even more guessing.
for best comparison compare the pal with ntsc like US, UK, FR, DE and IT. most of them should exist. pal has a better track record then NTSC or never twice the same colour.

I have an extensive collection of PAL region 4 and NTSC region 1 DVDs. For stuff that was converted from NTSC to PAL in the analogue domain in the previous century, I have found NTSC version is usually better in terms of colour and resolution. I put this down to the conversion equipment being poor in the previous century.

My S90D is calibrated with an i1d2 profiled off a cr-200. My 20p is ruler flat power law 2.4 and my saturation sweeps are all under DE 2000 2.0 including Pantone skin patches in HCFR. But more importantly my colour luminance through the sweep is stable, that was something that took me hundreds of hours to achieve.

[huhn]

you should really switch to itp on that screen...

[flossy_cake]

i have a calibrate display it's just orange juice in faces in all of the changed images.



So Picard's suit is supposed to be purple then? How do you explain it being red in the blu-rays ?

Also, and I mentioned this before, the OPs images are using a ffmpeg filter called zscale to perform the gamut conversion. Whereas I am using MadVRs pixel shader. The two don't necessarily result in the same colour depending on whether it uses "perceptual" or "relative" colourimetric intent

https://www.cambridgeincolour.com/tutorials/color-space-conversion.htm

After watching more of DS9 season 1 at 470M I am now convinced that whole season is also 470M.

The reason I was quick to judge is because for some reason whenever I saw an oversaturated skin tone I was weighting that more strongly than an undersaturated and greenish skin tone. At 170M, season 1 has far more "bad" skin tones when "green and pale" is given equal weight as "reddish and oversaturated".

Also, these old analogue masters are full of mastering errors, the colour is not even consistent from one episode to the next or from one scene to the next. So to find one scene that looks off and judge it based on that was wrong on my part.

[huhn]

So Picard's suit is supposed to be purple then? How do you explain it being red in the blu-rays ?

Also, and I mentioned this before, the OPs images are using a ffmpeg filter called zscale to perform the gamut conversion. Whereas I am using MadVRs pixel shader. The two don't necessarily result in the same colour depending on whether it uses "perceptual" or "relative" colourimetric intent

https://www.cambridgeincolour.com/tutorials/color-space-conversion.htm

After watching more of DS9 season 1 at 470M I am now convinced that whole season is also 470M.

The reason I was quick to judge is because for some reason whenever I saw an oversaturated skin tone I was weighting that more strongly than an undersaturated and greenish skin tone. At 170M, season 1 has far more "bad" skin tones when "green and pale" is given equal weight as "reddish and oversaturated".

Also, these old analogue masters are full of mastering errors, the colour is not even consistent from one episode to the next or from one scene to the next. So to find one scene that looks off and judge it based on that was wrong on my part.
the BD is also not orange face and i do not name the DVD to be just correct. it's just green puke as some would say here but the "correction" done here just result in orange faces...

and the reason i named pal dvd is not resolution or quality...
having one been done wrongly by a colorist is one thing if all look like "puke" it was most likely supposed to be "puke" that's why.

[nevcairiel]

Plus, if using MadVR we must remember to go in the settings and choose "calibration > this display is already calibrated to the following gamut /primaries > 709" (replace 709 with whatever your display's gamut is, obviously, but it's typically 709). Otherwise MadVR won't perform any gamut conversions

I was going to comment on that, its surprising how many PC video players just ignore primaries/gamut and just pass it through untouched to the screen, which is how you get old SD content looking wrong.

Even madVR doesn't by default just handle your screen as 709, which it really should, since thats been the SDR PC standard for ages.

[flossy_cake]

the BD is also not orange face and i do not name the DVD to be just correct. it's just green puke as some would say here but the "correction" done here just result in orange faces...
.

The thing is, I would say the colour of Picard's suit is a better reference than a random skin tone of a random scene.

Because that suit is not supposed to be purple, and Picard's face does not look like "orange juice" here either, so the 470M decoding on the right is 100% the correct way to decode the colours of that episode of the NTSC DVD in my opinion

https://i.imgur.com/SzYms0E.jpeg

[flossy_cake]

btw, I'm not claiming that every episode of every Star Trek series on NTSC DVD is 470M, which is something the OP seems to be suggesting, but there are definitely individual episodes/seasons that are 470M in my opinion. And in the case of TNG Season 1 there even appears to be individual scenes that flip flop between 170M/470M (mastering error)

[flossy_cake]

Also I think we are all a bit conditioned to thinking desaturated cyan skin tones are the norm cause that's how movies have been graded for the last 10 years

[huhn]

I was going to comment on that, its surprising how many PC video players just ignore primaries/gamut and just pass it through untouched to the screen, which is how you get old SD content looking wrong.

Even madVR doesn't by default just handle your screen as 709, which it really should, since that's been the SDR PC standard for ages.

yes easy mistake to make i ask to default to bt 709 gamma 2.4. people that care usually calibrate to that.
it also currently defaults internally to gamma 2.2 which can also be a problem in specific situations.
the fact that windows up to this day defaults to sRGB does not help... that's a very deep hole.

The thing is, I would say the colour of Picard's suit is a better reference than a random skin tone of a random scene.

Because that suit is not supposed to be purple, and Picard's face does not look like "orange juice" here either, so the 470M decoding on the right is 100% the correct way to decode the colours of that episode of the NTSC DVD in my opinion

https://i.imgur.com/SzYms0E.jpeg
no i will not say that but again no source image... it's just one scene is it purple in all of these there is maybe a sun there depending on it's white point it may made the skin orange or what ever that dvd image is supposed to be.

it is pretty much not possible that they randomly master one ep with m470 why would they even have a device calibrated to that.

you can easily make red purple with some lightning so it is not impossible.

[flossy_cake]

it is pretty much not possible that they randomly master one ep with m470 why would they even have a device calibrated to that.


Well I picked up those NTSC DVDs of Dr Quinn Medicine Woman which was made 1993-1998, same time as Deep Space Nine,
and they are flagged as 470M for season 1 up to S01E16 "The Operation"

https://i.imgur.com/VCrDd9O.png

Then at S01E17 "The Secret" it switches to 170M/SMPTE-C

https://i.imgur.com/IKIHYM7.png

This gamut metadata is only available when playing the disc itself with DVD Navigator filter in MPC-HC.
If I remux to mkv or open the vob, gamut metadata is not detected

If I was a videophile back then I probably would have used it too cause it's a wider gamut and preserves more of the colours in the celluloid

https://i.imgur.com/DgtS6q6.png


But the blue is different, I think this bit here explains why Dax's blue uniform and the blue glow under the runabout in the title screen is more teal:

https://i.imgur.com/8kv9t6p.png

[flossy_cake]

But, I think we can fix the blue issue by putting our display in 2020 gamut and telling MadVR "my monitor is calibrated to 2020"

But the teals are kind of magical somehow in that show, it's some weird psychological effect

[flossy_cake]

Well I played back some more of those Dr Quinn 470M flagged episodes at 470M and the skin tones are way off, glowing crazy orange and definitely 100% wrong.
So I think that 470M flag is a mastering error on the DVD of Dr Quinn.

If I play the disc on a Sony DVD player it doesn't treat it as 470M either, it's either treating it as 170M or just mapping it straight to 709
(I have no way of telling which, but it's definitely not interpreting it as 470M saturation levels)

edit: examples

https://i.imgur.com/zERNikH.png
https://i.imgur.com/Ui5Gog4.png
https://i.imgur.com/Lj9NiQy.png
https://i.imgur.com/yHmGaFI.png

[flossy_cake]

I think all we can really conclude at this point is ... there are mastering errors

But I would say if the content looks unusually undersaturated , AND the disc is flagged 470M, AND the colours look "less wrong" at 470M, THEN I will use 470M

Plus, if the hue seems off (green/magenta) that could be NTSC analogue phase shifting and to try adjusting hue 3 to 4 clicks either way to what looks better. After all, that's what the hue control was designed for

[SeeMoreDigital]

Star Trek Next Generation is currently being shown on Sky Mix HD here in the UK and the colours look very different to the original standard-def broadcasts. More like the images on the right...

[flossy_cake]

Star Trek Next Generation is currently being shown on Sky Mix HD here in the UK and the colours look very different to the original standard-def broadcasts. More like the images on the right...

Is it the HD remastered version? What colour is Picard's uniform - red or purple?

[flossy_cake]

470M has nice teals

https://imgsli.com/NDAwMTI0
https://imgsli.com/NDAwMTIz
https://imgsli.com/NDAwMTIy

(Voyager S01E11)

edit: lemao, on both my IPS screens there's not a very big difference between the two, while on my calibrated QDOLED there's a massive difference.

[SeeMoreDigital]

Is it the HD remastered version? What colour is Picard's uniform - red or purple?I think it must be. I'm actually watching episode 13 (11001001) from season 1 right now!

I guess the colour is more on the purple side as I've just viewed a holodeck scene with Riker, Picard and a woman wearing a red dress.

[flossy_cake]

I'm 100% convinced now that Season 1 of DS9 is encoded as 470M on the NTSC DVD release. Yes, there are some scenes in certain episodes where the skin tone is pushed, but these are not very common, and there are other episodes where the entire episode is good. If it was truly the wrong gamut, almost every scene would have glaring problems. I think the dull colours of 170M is actually hiding these colour grading fluctuations. An analogy might be if you use a very bluish white balance that also tends to hide colour issues.

[flossy_cake]

It's also possible there may be a better or optimal way of doing the conversion from 470M to 709. Apparently the SMPTE specification recommends a certain way of doing the conversion to match a given set of CRT phosphors. Apparently most CRTs back then weren't capable of the wide gamut of 470M and that's part of the reason why there are colour inconsistencies with it.

[Sunspark]

edit: lemao, on both my IPS screens there's not a very big difference between the two, while on my calibrated QDOLED there's a massive difference.

On my TN monitor, there is a massive difference. The 470M is strongly teal while the left side is a bit desaturated and bluish-purple. The 170M is probably more correct, but needs higher saturation.

That's strange that your IPS monitors aren't showing correctly.. do you have Windows colour management set up and also in the browser?

[flossy_cake]

On my TN monitor, there is a massive difference. The 470M is strongly teal while the left side is a bit desaturated and bluish-purple. The 170M is probably more correct, but needs higher saturation.

That's strange that your IPS monitors aren't showing correctly.. do you have Windows colour management set up and also in the browser?

Colour management is all disabled globally, on that IPS machine I only do vcgt cal and on the S90D it's all done at the display. I don't think my S90D has accurate blues

Edit: but if I crank the brightness up on the IPS then I notice more of a difference. So maybe IPS has some blue saturation issues at low luminance or something.