or Connect
AVS › AVS Forum › Display Devices › Digital Hi-End Projectors - $3,000+ USD MSRP › Time to define an alternative to ANSI contrast?
New Posts  All Forums:Forum Nav:

Time to define an alternative to ANSI contrast? - Page 2

post #31 of 341
Quote:
Originally Posted by ChrisWiggles View Post

Say, I think a minor but important thing to quantify is what the average average picture level is.

I.e. what *is* the average APL of most films over time? ANSI is 50% APL, but that's still way higher than I think most images. I don't think I've ever seen a good number or range for the average APL of most movie content. That's important to have an idea of so that you can whip up a test pattern at that APL or a few in the general range of average APL.

Doesn't a Kodak grey card which is 18% represent the average light and color an average image returns to a camera? Would that be a good starting point for an APL ANSI test?

Gary
post #32 of 341
Quote:
Originally Posted by Gary Lightfoot View Post

Doesn't a Kodak grey card which is 18% represent the average light and color an average image returns to a camera? Would that be a good starting point for an APL ANSI test?

Gary

No, an 18% luminance card is half gray. It's 18% luminance, or 50% gray. It's actually 50% APL because it's a full field of 50% gray. I think most scenes are a lot lower than 50% APL.
post #33 of 341
If I go from a Ruby with DI to a native 15,000:1 pj (RS1), it seems that light reflection control in the room would become even more important with the RS1. I think this would be the case because the dynamic range will be much greater.

For instance, with a star field and a bright object - before the the Ruby the bright object may have only been half as bright as it should have been. Now with the RS1 the bright object will be much brighter, but the darks will be similar as with the Ruby. So in that case it seems logically then that there is much more light causing reflections in medium to low APL scenes and therefore would have a greater impact on the ANSI CR.

Is this line of thinking correct?
post #34 of 341
Companding Calculator
http://www.brucelindbloom.com/index....dCalcHelp.html
*Bottom of page click [Back to Calculator]

Quote:


A companding function is, in general, a nonlinear transformation of luminance (Y). It is often used to redistribute luminance values in a manner that is more uniform to the eye, in preparation for quantizing to a fixed number of discrete levels (e.g. 256 levels using an unsigned, 8-bit integer). This calculator supports conversions among luminance, CIE L*, density and gamma functions.

L*50 = 18%

L* 20 = 2.9%
post #35 of 341
Quote:
Originally Posted by ChrisWiggles View Post

No, an 18% luminance card is half gray. It's 18% luminance, or 50% gray.

To be nit-picky, an 18% card shouldn't be called "50% gray" because that's a mathmatical expression. To the average eye, a gray swatch that reflects 18% of the light incident upon it is considered "medium gray", but that is purely a psychological response. (Yeah I know, another word for "medium" is "50%", but "medium" indicates a lack a precision that is more indicative of our less than perfect understanding of "typical human perception".)

When you video record a 18% card, you will get an APL of approximately 42% depending on the gamma of the video camera (the 2.2 gamma of REC709 actually works out to almost 2.0 exactly when the specified tail is taken into effect.)

And finally when you show the 42% APL signal on a monitor it will likely output a luminous of less than 18%. (Typically 11% or so for a CRT due to the higher gamma of the display)

All of this is why I strongly recommend that any ANSI-esque pattern developed should be a mixture of white and black to eliminate the role of display gamma in the measurement.

-Mr. Wigggles
post #36 of 341
Sure, I probably should have said 50% lightness=18% luminance. I was loosely assuming that when I said "50% gray" that it was perceptual lightness rather than physical luminance. Sorry if that was unclear.
post #37 of 341
Quote:
Originally Posted by MrWigggles View Post

To be nit-picky, an 18% card shouldn't be called "50% gray" because that's a mathmatical expression.

http://www.brucelindbloom.com/index....dCalcHelp.html
Quote:


You can learn things from this graph. For example, selecting Y as the input and L* as the output, you get the curve that maps the uniform intensity (i.e. energy) scale to the uniform perceptual scale. From this, you can see that a middle visual gray (L* = 50) corresponds to a luminance (or reflectance) of 0.18 (or 18%)

What value for gamma gives a companding function that most closely represents the CIE L* function (i.e. a uniform perceptual scale)? You can see that a gamma value of 2.2 is not too bad of a compromise because it roughly follows the diagonal:

FWIW, I have done research in regard to the limits of the CIE perceptual model, i.e. LDR 8-bit images, and I've found evidence presented in the form of a master thesis that states " L* value of '8' is the effective break point where this model breaks down". Corresponding to max contrast ratio of 112 or a density 2.05.
post #38 of 341
Quote:
Originally Posted by darinp2 View Post

It looks like that 5% APL pattern you were referring to has 5% of the screen covered by white rectangles. If I were to estimate the value for that image in that calculator I would change the percentage of the screen covered by the checkerboard to 10% (twice the 5% because the checkerboard contains equal amounts of white and black). And I would change the room gain to 0 to just consider the projector. As an example, for something with 800:1 ANSI CR and 6000:1 on/off CR and those other choices, that calculator would estimate the CR in that image at about 3600:1.

Let's try another example, but taking the room into account this time. Let's suppose the room limits ANSI CR on the screen to about 200:1, so the reflectivity value is 0.01.

According to the calculator, a projector with 4000:1 on/off CR and 250:1 ANSI CR would produce 884:1 CR for that pattern.

If we boost the on/off CR to 8000:1 but keep the 250:1 ANSI CR, the pattern's CR would be 982:1, which is an increase of 11%.

Or, instead, if we boost the ANSI CR to 850:1 but keep the 4000:1 on/off CR, the pattern's CR would be 1183:1, which is an increase of 34%.

The 34% increase is about 3 times as much as the 11% increase. Basically, the ANSI CR boost would improve the pattern's CR more than the on/off CR boost would, even though the pattern has a low APL, and even though the room is reflective.

Just for grins, if we boost both at the same time (on/off CR to 8000:1 and ANSI CR to 850:1), the pattern's CR would be 1366:1, which is an increase of 55%.

In addition, even if we boost the on/off CR to infinity but keep the 250:1 ANSI CR, the pattern's CR would be 1105:1, which is still less than what the ANSI CR boost produces.
post #39 of 341
Quote:
Originally Posted by Erik Garci View Post

Let's try another example, but taking the room into account this time. Let's suppose the room limits ANSI CR on the screen to about 200:1, so the reflectivity value is 0.01.

According to the calculator, a projector with 4000:1 on/off CR and 250:1 ANSI CR would produce 884:1 CR for that pattern.

If we boost the on/off CR to 8000:1 but keep the 250:1 ANSI CR, the pattern's CR would be 982:1, which is an increase of 11%.

Or, instead, if we boost the ANSI CR to 850:1 but keep the 4000:1 on/off CR, the pattern's CR would be 1183:1, which is an increase of 34%.

The 34% increase is about 3 times as much as the 11% increase. Basically, the ANSI CR boost would improve the pattern's CR more than the on/off CR boost would, even though the pattern has a low APL, and even though the room is reflective.

Just for grins, if we boost both at the same time (on/off CR to 8000:1 and ANSI CR to 850:1), the pattern's CR would be 1366:1, which is an increase of 55%.

What are you concluding from this?
post #40 of 341
Quote:
Originally Posted by lovingdvd View Post

What are you concluding from this?

It shows that ANSI CR matters, possibly more than on/off CR matters. Even if the scene has a low APL, ANSI CR matters, possibly more than on/off CR matters. Even if the room is reflective, ANSI CR matters, possibly more than on/off matters.
post #41 of 341
Quote:
Originally Posted by Erik Garci View Post

It shows that ANSI CR matters, possibly more than on/off CR matters. Even if the scene has a low APL, ANSI CR matters, possibly more than on/off CR matters. Even if the room is reflective, ANSI CR matters, possibly more than on/off matters.

I don't think that conclusion follows from your example. The choice to go to an ANSI increase more than 2x to an unrealistic ANSI CR measure of 850:1 is dubious.

Instead, if you double ANSI (as you doubled on/off CR previously) in this particular example, you end up with a CR of 1077, which is a 21% increase which is still twice the 11% with a coublind of on/off CR in this particular example.
post #42 of 341
Quote:
Originally Posted by ChrisWiggles View Post

I don't think that conclusion follows from your example. The choice to go to an ANSI increase more than 2x to an unrealistic ANSI CR measure of 850:1 is dubious.

Greg Rogers recently measured 845:1 modified ANSI CR on the Sharp 20K. If 845:1 is realistic, then isn't 850:1 realistic also?

Anyway, I'm not quite sure what you are taking issue with. Are you saying it is wrong to conclude that ANSI CR can matter more than on/off CR in my example?
post #43 of 341
Quote:
Originally Posted by ChrisWiggles View Post

Instead, if you double ANSI (as you doubled on/off CR previously) in this particular example, you end up with a CR of 1077, which is a 21% increase which is still twice the 11% with a coublind of on/off CR in this particular example.

True, and doubling ANSI CR to 500:1 would still be better than going to 37,000:1 on/off CR instead in this example.
post #44 of 341
Quote:
Originally Posted by Erik Garci View Post

Greg Rogers recently measured 845:1 modified ANSI CR on the Sharp 20K. If 845:1 is realistic, then isn't 850:1 realistic also?

Anyway, I'm not quite sure what you are taking issue with. Are you saying it is wrong to conclude that ANSI CR can matter more than on/off CR in my example?

I was not aware of that. What I was taking issue with at first was the assertion of the supremacy of ANSI over on/off CR. I was going to write a long response but I got distracted with other things, and I don't really take issue with what you said anyway. I don't really have any issues with what you said, I just am not convinced that the particular example is fully convincing, but I don't necessarily disagree either. Of maybe to put it another way: I don't really know what I meant.
post #45 of 341
Quote:
Originally Posted by ChrisWiggles View Post

I don't really have any issues with what you said, I just am not convinced that the particular example is fully convincing, but I don't necessarily disagree either.

I had appended the following sentence to my post to try to make it more convincing.
Quote:
Originally Posted by Erik Garci View Post

In addition, even if we boost the on/off CR to infinity but keep the 250:1 ANSI CR, the pattern's CR would be 1105:1, which is still less than what the ANSI CR boost produces.

In other words, boosting ANSI CR from 250:1 to 850:1 is better than boosting on/off CR from 4000:1 to any amount, even to infinity, in this example. Is that convincing enough? I'm not sure if I can make it more convincing than that.
post #46 of 341
Quote:
Originally Posted by ChrisWiggles View Post

I was not aware of that. What I was taking issue with at first was the assertion of the supremacy of ANSI over on/off CR.

I think Erik was just saying that it ANSI CR can be more important even in an image that is closer to absolute black than the ANSI CR pattern, but understands that in other cases on/off CR would be more important than ANSI CR. Even with the same pattern. As we both know, either one can be the weak link. Start with very low on/off CR and okay ANSI CR and the on/off CR is that one that basically needs improvement.

And it can be even more complicated than the numbers since improving the places where a projector is weak is usually more important than improving the places where it is strong, even if the percentage improvements are the same. That is one of the things I tried to emphasis in this section of my article on CR:

http://www.hometheaterhifi.com/volum...06-part-5.html

--Darin
post #47 of 341
Quote:
Originally Posted by darinp2 View Post

As we both know, either one can be the weak link. Start with very low on/off CR and okay ANSI CR and the on/off CR is that one that basically needs improvement.

Regarding this particular example (the test pattern and the room), we can determine which spec is the weaker link (on/off or ANSI) by comparing them to each other as follows. If the on/off is greater than 9 times the ANSI, then ANSI is the weaker link, in the sense that doubling ANSI would improve the pattern's CR more than doubling on/off would. Or, if the on/off is less than 9 times the ANSI, then on/off is the weaker link, in the sense that doubling on/off would improve the pattern's CR more than doubling ANSI would.

For example, if the projector has 4000:1 on/off and 300:1 ANSI, then ANSI is the weaker link, since 4000 > 9 * 300. Or, if the projector has 3000:1 on/off and 400:1 ANSI, then on/off is the weaker link, since 3000 < 9 * 400.

Furthermore, if the on/off is greater than 18 times the ANSI, then ANSI is so much the weaker link that doubling ANSI would improve the pattern's CR more than going to infinite on/off would. For example, if the projector has 4000:1 on/off and 200:1 ANSI, then doubling ANSI would improve the pattern's CR more than going to infinite on/off, since 4000 > 18 * 200.
post #48 of 341
Thread Starter 
Quote:
Originally Posted by Erik Garci View Post

Darin's explanation is correct on how to use the calculator.

As for the test pattern on page 15 of the document, it is labeled "5% luminance, 26% APL grey." I measured the approximate number of white and black pixels, and I confirmed that 5% are white, and 95% are black.

The average luminance of this test pattern varies somewhat depending on the display. On a hypothetical display with infinite on/off CR and infinite ANSI CR, the average luminance would be 5% (relative to 100% for a "full on" test pattern).

The APL of this test pattern is 5% (not 26%), which can be calculated as follows:
(5%*white) + (95%*black) = (5%*100%) + (95%*0%) = 5%

Let's consider a different test pattern, one that is a solid shade of gray with a signal level of 26%. It has an APL of 26%, obviously. If it is displayed with a gamma of 2.22 on the hypothetical display, then its average luminance would be 5%, since 0.26^2.22=0.05.

Basically, the "26% APL grey" in the label refers to a solid gray test pattern, which would have the same 5% average luminance.

Hi Erik, sorry I'm behind in my reading. So let me see if I understand this correctly. The average luminance with a full white (100%) and full black (0%) is easily calculated and yields some value let's call it X, what they then do in the white paper is work backwards and figure out what APL is needed for a solid grey test pattern with a 2.22 gamma which yields the X average luminance. Wow, this explains it but I would never have guessed this relationship if you hadn't pointed it out. Thanks!
post #49 of 341
Thread Starter 
Quote:
Originally Posted by MrWigggles View Post

To get the square as far away from the other corner as possible.

Also if you put it in the center, a projector with poor uniformity might actually score higher. By using opposing corners, you will be using parts of the projected image that are nominally the same brightness.

Some technologies are much better with light rolloff on the edges than others. CRTs for example are brighter in the center than the edges. This begs the question then on how to create a test pattern without making brightness uniformity the dominant factor. Maybe moving the white boxes to the quadrant between the center and extreme edge will be best. Thoughts?

Quote:
You could also do an ANSI checkerboard version of my image by starting with a black background and adding eight little white squares surround - thus keeping the image still 99% black. But I wanted the square to be large enough in case you need to make the measurement close to the projector in situations where your light meter doesn't have enough sensitivity to measure the black corner well.

This is a good point of practical consideration. The number of white boxes has to be limited to keep the area large enough to cover a light sensor. I think 8 white boxes is going to be a problem, but perhaps using 4 (or maybe 5 with one in the center) will be okay.
post #50 of 341
Thread Starter 
Quote:
Originally Posted by Erik Garci View Post

It shows that ANSI CR matters, possibly more than on/off CR matters. Even if the scene has a low APL, ANSI CR matters, possibly more than on/off CR matters. Even if the room is reflective, ANSI CR matters, possibly more than on/off matters.

This is an interesting relationship that I hadn't expected. CRT for example has great on/off CR but limited ANSI and yet the image it yields has a lot of depth.

It's also worth pointing out that this relationship is derived from your model rather than from actual data. At the very least, creating a low APL test pattern will help to validate the model. I would think that things would work out as expected in mid to high APL scenes but in low APL scenes there seems to be so much going on with dynamic gamma, dynamic irises, etc. that having actual measured values would be beneficial.
post #51 of 341
Quote:
Originally Posted by Mark Petersen View Post

This is a good point of practical consideration. The number of white boxes has to be limited to keep the area large enough to cover a light sensor. I think 8 white boxes is going to be a problem, but perhaps using 4 (or maybe 5 with one in the center) will be okay.

You shouldn't limit yourself to thinking only about squares. One of the reasons that the old test patterns were checkerboards was because it was easier to implement the signal generator hardware. But since these patterns are probably going to be photoshopped and then encoded, we are no longer limited to what shapes are easy to implement in hardware.

Not that I have any better ideas than using squares, I just want to make sure we don't limit ourselves unnecessarily.
post #52 of 341
Quote:
Originally Posted by Mac The Knife View Post

You shouldn't limit yourself to thinking only about squares. One of the reasons that the old test patterns were checkerboards was because it was easier to implement the signal generator hardware. But since these patterns are probably going to be photoshopped and then encoded, we are no longer limited to what shapes are easy to implement in hardware.

Not that I have any better ideas than using squares, I just want to make sure we don't limit ourselves unnecessarily.

In other words, don't be a square!
post #53 of 341
Quote:
Originally Posted by Mark Petersen View Post

As I recall there was a long thread that discussed this calculator can you find a link to it?

I found an archived thread that discussed it, but it's a short thread.

Instantaneous CR from on/off CR and ANSI CR
post #54 of 341
Thread Starter 
Quote:
Originally Posted by Erik Garci View Post

I found an archived thread that discussed it, but it's a short thread.

Instantaneous CR from on/off CR and ANSI CR

Excellent, thanks!

Btw, I'd like to go to the next step in the current discussion and actually make some test patterns. I'd like to hear from people what luminance percentage that they would like to see that matches a typical low, mid and high APL scene. I was thinking of 1%, 5% and 10% luminance patterns (black background with increasing areas of full white). But after thinking about it, even these patterns might be a little too bright to truly represent a typical low, mid and high APL film scene.
post #55 of 341
Quote:
Originally Posted by Mark Petersen View Post

Excellent, thanks!

Btw, I'd like to go to the next step in the current discussion and actually make some test patterns. I'd like to hear from people what luminance percentage that they would like to see that matches a typical low, mid and high APL scene. I was thinking of 1%, 5% and 10% luminance patterns (black background with increasing areas of full white). But after thinking about it, even these patterns might be a little too bright to truly represent a typical low, mid and high APL film scene.

Once you have the test patterns perhaps you can hook up with GetGray to have the patterns included on his calibration disc.
post #56 of 341
Thread Starter 
Quote:
Originally Posted by lovingdvd View Post

Once you have the test patterns perhaps you can hook up with GetGray to have the patterns included on his calibration disc.

GetGray is a good idea. I was also thinking of asking dr1394 to include it in his HD-DVD test pattern disk: http://www.w6rz.net/
post #57 of 341
Quote:
Originally Posted by Mark Petersen View Post

GetGray is a good idea. I was also thinking of asking dr1394 to include it in his HD-DVD test pattern disk: http://www.w6rz.net/

Sounds good. One thing that will be interesting is that as we experiment with these patterns we can measure our pjs and share our results here to get a sense for how different pjs are performing. This should also tell us if the patterns seem to be a good test indicative of each pjs capabilities with regards to ANSI CR.
post #58 of 341
Thread Starter 
Quote:
Originally Posted by lovingdvd View Post

Sounds good. One thing that will be interesting is that as we experiment with these patterns we can measure our pjs and share our results here to get a sense for how different pjs are performing. This should also tell us if the patterns seem to be a good test indicative of each pjs capabilities with regards to ANSI CR.

Exactly. It'll also be interesting to see how different projectors with different DI algorithms fair in the low APL test pattern too.
post #59 of 341
One thing that is very interesting to see using the Contrast Calculator is how much on/off CR plays a role for ANSI CR at lower APLs.

I'd like to make sure I am understanding something correctly about this though...

In the table of results in the calculator it shows the various % of white. Which one of these values is most representative of what we actually see in most typical scenes?

Its been said here that the typical APL in scenes is about 20-ish. Does this mean that the results at 10, 20% and 50% are more important that what it shows at 100%?

Also on somewhat of a different note - is it a safe assumption to say that generally speaking if you have a pj with very good ANSI CR and light/reflection control in the room that you'd want to aim for a gamma of about 1.5 (as compared to a poorer ANSI CR pj and/or poor light conditions where you may want to aim for 2.2)?
post #60 of 341
Quote:
Originally Posted by lovingdvd View Post

In the table of results in the calculator it shows the various % of white. Which one of these values is most representative of what we actually see in most typical scenes?

Its been said here that the typical APL in scenes is about 20-ish. Does this mean that the results at 10, 20% and 50% are more important that what it shows at 100%?

In another thread, I calculated the APL of a scene from the movie Gladiator to be around 15%.

To calculate the contrast of this scene approximately, enter these values:
2.5 gamma
5% screen area covered by the checkerboard
90% checkerboard stimulus

Then enter your own values for on/off CR, ANSI CR, and room gain.
Quote:
Originally Posted by lovingdvd View Post

Also on somewhat of a different note - is it a safe assumption to say that generally speaking if you have a pj with very good ANSI CR and light/reflection control in the room that you'd want to aim for a gamma of about 1.5 (as compared to a poorer ANSI CR pj and/or poor light conditions where you may want to aim for 2.2)?

1.5 is way too low for a display gamma. Did you mean 2.5?
New Posts  All Forums:Forum Nav:
  Return Home
AVS › AVS Forum › Display Devices › Digital Hi-End Projectors - $3,000+ USD MSRP › Time to define an alternative to ANSI contrast?