Setting contrast with Spears and Munsil disc - Page 13

I've seen numbers above 235 on pretty much everything I've looked at.

I'm not convinced that the numbers relate to actual or useful picture content for BD, however rather than draw conclusions I'm trying to figure that particular issue out.

Broadcast seems to have useful information above 235 , again however I have not satisfied myself against the possibilty that its not in fact just "noise".

Visually BD in most examples I've looked at seems to look fine with a hard clip at 235. Broadcast I'm less convinced.

But unless you can clip to 235 in RGB space, you still need to calibrate to 254, because you never know what kind of legal YCC value might transform to an RGB above 235.

I think we should concentrate on what is actually displayed on the screen regardless.

Hanging on to variation till 254 is a waste if all that exists above 235 is garbage.

I've yet to find a solid example of BD that looks notionally compromised when mapping peak white to 235 and believe me I've been looking. Every time I look at imagery I'm convinced will show detail above 235 I see at best a handful of isolated pixels and what looks like edge ringing from resampling.

Again I say ignore what you read from established names , ignore the evidence on the net that backs up a particular rationale. LOOK at the imagery , use John's script to isolate candidate images and find real examples.

Unless we do this at least half of us will be failing in optimising our displays with one of the most basic user controls.

I'm saying that the question wasn't asked correctly. I'm well aware of Charles Poynton and every other name you've mentioned. ISF is a joke so let's not sully the credentials of those that have them with that nonsense. I've read completely foolish and entirely unsupportable things by some of the people you've listed (naturally Poynton brings rigor to the discussion). Many of them about this very subject.

What I'm saying is that you can speculate about things or you can measure them and draw careful conclusions. Of all the names you've mentioned I'm only aware of Stacey and Don looking at on disc data. However they only observe that it exists they didn't characterize it. The fact that it exists on the disc is only the beginning of understanding. E.g. was on it on the DI? Is it surrounded by sufficient dark pixels that it's perceptible? Can you discern it in an A/B blink test?

After you determine that above ref. data even matters then you can address the limited dynamic range of any given display and decide if throwing away 15%* of your luminance for .001%** of the pixels is even worth discussing.

*handwaving
**more handwaving

I don't understand how you guys think you can clip data and have it disappear?

if you start clipping you get a ragged top to your display as R, G and B will clip at different points.

If you can clip in RGB space, fine, whatever, clip if you want.
But that is not 99% of displays.

The YCC to RGB conversion happens, and when it does you can't control if the value goes over 235 or not.

I'm pretty sure nobody has been suggesting doing anything to the YCrCb data, certainly not clipping it. My understanding is that Mr.D is referring to 8bit Studio RGB values when he talks about clipping and also this is clipping within the display contrast control. That he happens to be doing that clipping prior to he display is to some extent irrelevant to the wider discussion.

John

I think the question is now not so much are there pixels > 235 in studio RGB. The question is now are those pixels valuable if so how valuable are they. There is not much evidence yet supporting the idea that there is anything particularly valuable above 235 for BD.

John

It would be great to find out what actually happens. Would it also be possible to clarify the approval process for the various formats. Is the grading step the last time material changes are made to a typical master, or are significant manual adjustments made later in the process?

John

I would dispute this to a certain degree ( obviously I do not have hands on experience of every display out there). Out of all the displays I've owned the only one that didn't have a contrast control that was essentially a white clip control is my current plasma (42PHD8). ( which has an input level control that essentially controls clipping)

All the others I've owned (including my HD1) have had predictable contrast control behaviour. (the higher it goes the lower it clips).

There will always be displays that have limitations/ foibles. Obviously in those cases you have to either make the best of it or replace it with one that is less problematic.

I don't think we should get too side tracked though all we are interested in is where useful detail stops on material at the moment. The ramifications are something we should discuss after we've pulled through enough data to reach a consensus.

From what I can gather (and this is by no means concrete yet).

Film negative gets scanned to 10bit log. This is then graded with reference to a print color model ( profiled displays 3d luts of desired end print stock).

DCi and video versions then created with conversion luts that convert from the print stock to desired standard target.

Thats the hypothetical perfect model.

Real world what happens is that the print doesn't always nicely fit into the video range and some compromises have to be subjectively decided upon.


---

However I have come across some differing methods: (excuse the rambling)

Some places convert to linear right at the start and view with suitable LUTs to give them an end print environment for linear material . This makes zero difference to the first method assuming that the colorist knows what colorspace they are operating in (in my experience companies that enforce linear workflow tend to do it to prevent the operator from having to get involved in colorspace considerations rather than for any immediate quality improvement).

I regularly come across people who tell me you can't grade in log and it just makes me laugh: you need to know what the pros and cons of each working space are and move in and out accordingly. I also regularly come across work from companies that have enforced linear workflow and find that the work has a 10 code value rounding error as a result of imprecise conversion and or too many multiple conversions inducing rounding errors. You could argue that 10 code values is nothing... not sure a good colorist would agree with that.

I've come across one place that uses rec.709 monitors and a colorist there told me that to convert the film to video they just "take the film lut off to see what it looks like as video". Which is either a massive oversimplification of the process or the guy didn't really know what was going on. Interestingly enough that company had the DI pulled a few months into production.

I also had a colorist talk to me about "some sort of linear video format for final delivery" which is of course an oxymoron. They also used the terms "2k" and 1080p interchangably and managed to badly clip some material (and I mean bad!)

To be honest you could take any colorspace as a working environment as long as you were able to move to any others that were required and still get the desired controlled result. Some people may work in a limited video environment like rec.709 with float precision (maintaining all the data from the scan) and then bodge a conversion to whatever they need but I kinda doubt it as you are then never appraising with the largest set of visual precision regardless of the data you are shuttling about.

I work log input with print lut and move into linear and back out when I need to for specific types of operations. I reference it all back in log as print though. I also linearise video as and when required for the same reasons and I hardly ever see other operators doing this ( quite a few still seem to think video is linear anyway purely because its not log )

What I/we need to find out is where they are shooting for white when they convert to video regardless and unfortunately a lot of the people you would think might know this either don't or will answer back 235 or 254 without explicitly knowing that for a fact.

Is this done as simple brightness/contrast type adjustment or a more complex s-curve type of control?

John

I'm not sure I fully understand the process/terminology here. Does this mean that when grading is done you are looking at

1. The raw signal sent to the film printing process
2. What the film print should look like
3. What the projected film should look like on screen

Thanks

John

Arbitrary curves. The actual grading process is even more segmented than that although much more than simple keying and garbage mattes requires the use of a VFX type process to isolate the areas and color correct them without creating artifacts: ie me.

Old style lab grading used to be an additive offset process: a stop equates to 96 code values , 8 printer lights to a stop. You can actually very easily dial in stops in terms of code values on 10bit log material to replicate this type of grade with reasonable accuracy.

Nowadays the film is really just treated as a captured intensity range , it looks correct once its been graded to look correct in other words.

Probably a poorly worded question, I was asking about how any post-grading squash to video stage might operate. Or is this effectively another full grading type process.

John

All three at the same time!

The scans are "full negative density" rather than print density. So the data essentially behaves like the negative would.

A print is "high contrast" ie its less information than exists on the neg.

The display environment is lutted to make the data on the scan (negative) look like it would if you struck a "one light" print off it and then viewed the print in a projected environment.

So you are looking at print simulation derived from data that correlates to the negative.

In the same way you could look at video as your end target simulation but you still have the entire data from the negative available for manipulation. So even though you are targeting a smaller range you have the entire range from the negative to pick from. (hence the need for a little subjective compromise)

As to the way its graded: think parametric RGBHSI curves and if thats not enough they isolate the image by keying in terms of RGBHSI and every mathematical combination thereof.

And if that fails they manually saw the image apart with hand drawn roto.

And if the detail isn't there in the first place they can composite it into the shot although now we are really into something thats more like a visual effects pipeline. (I'm regularly called upon to do this sort of stuff where the DI guys start to struggle)

It can be a full regrade from the negative data: for example the case of older films that are being remastered just to video.

Its more usually a bit of a standard transform from the log grade then some touch ups on problematic scenes when its a contemporary film

FWIW, ringing, produced by scaling or sharpening algorithms, is not the only way to produce "garbage" BTB/WTW pixels. Chroma sub/upsampling can do that, too. Just imagine alternating black and red lines. If you subsample and later upsample the chroma, the black lines will have chroma information, which will result in negative values for Green and Blue. See more details here:

http://www.glennchan.info/articles/t...ma/chroma1.htm

This is true and I generally regard chroma upsampling issues as being generally the same as normal image scaling.

This is one area I'm looking into to see if some ( if not all) of the information we are seeing above 235 is actually just ringing and or compression artifacts.

I can simulate YCrCb and chroma subsampling issues all the way from 2k log to component 4:2:0 video ( think I did it once already). Compression issues I can't readily replicate but compression artifacts seem to be quite obvious in terms of false detail.

Ok, so based on what we know up to this day, for BD/DVD, top white at 235 is recommended. Isn't it?
In the other hand, what happens with mkv files (720p) available in internet that are generated from TV Shows (CSI, Lost, Heroes, The Pacific...)? Is top white recommended at 235 too for them?

We're not currently recommending anything we are making observations with a view to coming to some sort of consensus.

Illegal downloads are not particularly relevant to this discussion. You may as well ask how long is a piece of string.

Image scaling issues are very different to the chroma upsampling issue I just mentioned. Image scaling issues depend solely on the scaling algorithm chosen. E.g. with Sinc filtering you get a sharp picture with no aliasing, but tons of ringing. If you use Mitchell filtering instead, you get a noticeably softer image with some aliasing, but virtually no ringing. Or if you use Nearest Neighbor, you get extreme aliasing, but not a trace of ringing. So, scaling artifacts totally depend on which scaling algorithm you're using. With the chroma problem I mentioned, it doesn't matter *at all* which scaling algorithm you're using to upsample the chroma. It's not ringing. The same issue occurs with all algorithms, from Nearest Neighbor to Sinc.

The chroma artifact is not the result of scaling algorithm artifacts, instead it's the result of a mismatch between true luma/chroma resolution, which in certain situations produces "illegal" luma/chroma combinations. This effect usually happens on edges where a very dark (or bright) area meets another area which is strongly colored with only one of the 3 RGB components.


... and or chroma sub/upsampling artifacts.

Good point, a lot of the extreme WTW I've seen in my tests happens near things like tree/sky transitions where there will be chroma and luma jumps.

John

Ah interesting. Thanks.

I was reading it as just a manifestation of scaling artifacts from chroma upsampling propagated into RGB.

I was going to start with a 2k film scan ( maybe a 4k one if I have it handy) and see if I can create an image that exhibits some of the characteristic WTW features we seem to be seeing but are there any specific pattern that would be good to show this up?

Yeah. Let's say the sky is full white. By definition full white is not "colored". So Cb and Cr should 0 (if you interpret them as a signed int). Now imagine a full white sky pixel gets the green chroma information from a tree leaf. Suddenly we have Y to the max and green to the max. Let's do the math:

I hope my math is ok. If not, please someone correct me...

P.S: Of course in the real world the above example would only occur with a stupid box downsampled chroma channel. If at least linear filtering was used, the chroma of the edge pixel between sky and tree should get some kind of blended chroma, e.g. 50% chroma value of the sky and 50% chroma value of the tree. But still that can produce very big WTW values.

For BTB it's easy: Pick any movie which has red fonts on black background in the start/end credits. E.g. Highlander, or 7 Pounds.

For WTW, as I said before:

> This effect usually happens on edges where a very dark (or bright) area
> meets another area which is strongly colored with only one of the 3 RGB
> components.

So look for scenes where a bright white image area (Y very high, CbCr near to 0) edges to a pure red, green or blue image area (Y doesn't matter, CbCr as high as possible).

So this could explain the bulk of the image content above 235 in the almost famous example on the previous page? Especially as it doesn't seem to correlate with the photographic properties of the image.

Which famous example do you mean? Checked the previous page and there were a couple of image linked, but none stood out as famous for me.

IMHO:

(1) If WTW pixels are directly on an edge (especially if it's a bright white area edging to a pure colored area), then the WTW pixels were likely introduced by the chroma resolution mismatch problem.

(2) If WTW pixels are near edges, like ringing, then it's likely to be a scaling artifact. (Although in this case I think already the Y' channel should be too high).

(3) If WTW pixels are nowhere near an edge, it's like to be either a compression artifact or true WTW information.

Basically I'd consider all WTW pixels, which are limited to the area near edges, as probably being "bad".

"Almost Famous" picture of a kid walking towards a car with a big flare off the bonnet. There is a version with everything at 235 and below clamped to black. (page 11 of the thread)

The WTW remaining seems to live mainly along edges and brightness transitions. Photographically the flare is 235 in the dead centre which would make sense as the brightest region in the scene however there are some regions away from the centre of the flare that are above 235 but congregated around an edge.

Photographically you would expect this region to fall off in brightness which it does until it bangs off some edge detail

I think they're relevant, as they're used by a lot of people who want the best possible PQ with them too.

Anyway, I think there's no big deal in answering if that "illegal" downloads fall in the same box as Blu-Ray, broadcast, or they're a different case regarding recommended top-white level. Can you just tell us that, please?

Should behave as content in off line broadcast category.

John