# Contrast ratios meaningless?

Discussion in 'Displays' started by Mel W, Jul 16, 2003.

1. ### Mel W Auditioning

Joined:
Dec 4, 2002
Messages:
3
0
Contrast Ratio, what is it, how is it computed and does it really tell us anything of value?

This topic as mentioned in many recent reviews and equipment promotions has been bothering me for awhile now, so I thought I'd do some reading to find out what Contrast Ratio (CR) is. After you wade through all of the theory, contrast ratio is fairly easy to understand. Basically contrast ratio is another way to define the dynamic range of the difference in brightness of light. As in sound recordings, the larger the dynamic range the more realistic the sound, the same holds true for light. The larger the dynamic range (CR) of light the more realistic the picture.

Let's break Contrast Ratio into its component words and look at each. Contrast is the difference between light and dark items whether they are paint or light, in our video world mainly light. Ratio is the mathematical comparison of the difference of two similar measurements. The typical expression of this comparison is normally in the form of X:1 or something-to-one. So therefore CR would be the expressed difference between white and black (light and dark) in our images.

To derive the value of X for our expression of something-to-one, we use the formula C = Y1/Y2, where C is contrast, Y1 is our luminance measurement at the white point and Y2 is our measurement of black. Now the first step is to determine just how bright white (Y1) is (usually pretty easy) and how dark black (Y2) is (not so easy). By definition black is the absence of any light. This causes a problem in the formula as dividing anything by zero yields an undefined answer. Well that doesn't give us a valid ratio. So what do we do? Apparently we need to assign some value other than zero to the Y2 factor. But what should this value be and how large or small a value? Let's look at the recommended ways to measure contrast for electronic display devices, full on-full off and checkerboard. Well full on-full off seems easy enough: you take a reading of the set with it off and then with it on displaying a maximum white field. With the set off, obviously there is no light being produced so our value would be zero no matter how sensitive our measurement equipment is. That won't work as we already know that a ratio can't be found by dividing by zero. So what about the checkerboard? This pattern displays alternating areas of black and white information on the display. Since we know that all machines do not perform exactly to theory, then we know that there will be some small amount of light produced in the black area since the unit is operating. The value derived for black will depend upon how sensitive our measuring equipment is and how close to theory the device works.

For computational arguments sake, let's use some values representative of the calibration industry. For most calibrators, they use a solid state detector that reads luminance accurately down to 0.1ftL (foot Lamberts). A few "Top Gun" calibrators use a meter capable of reading down to .01ftL. Any light lower than that and the instruments return either a bad number or zero. Well we don't want to use bad numbers, and we can't use zero, so let's use the lowest known accurate point of luminance detection. For comparison we'll do the same thing for a typical lab grade meter, the cut off point of which is 0.001ftL. Let's look at some computations using these figures with readings from typical installations and with the meters reading the same display. (Results rounded for clarity)

----------------------------------------------------------------------------------------------------|
------------------|Contrast Ratio (X:1) for display device set to--|-------------------------|
------------------|100IRE with typical light output.-----------------|-------------------------|
------------------|-------------------------------------------------------|-------------------------|
Measurement--| Front------| Rear------| Direct View-| Plasma---|Examples of-----------|
device low-----| Projector--| Projector-| output in---| output in-|typical------------------|
end cut off-----| output in-| output in-| ftL 34-------| ftL 55-----|equipment-------------|
-----------------| ftL 12------| ftL 21----| --------------| ------------|used--------------------|
-----------------|--------------|------------|--------------|-------------|--------------------------|
0.16------------| 75----------| 131-------| 213---------| 344--------|Sencore 5000,---------|
-----------------| -------------| -----------| --------------| ------------|Progressive------------|
-----------------| -------------| -----------| --------------| ------------|Labs CA1-SE-----------|
-----------------| -------------| -----------| --------------| ------------|etc.---------------------|
-----------------|--------------|------------|---------------|------------|--------------------------|
0.01------------| 1200-------| 2,100-----| 3,400-------| 5,500-----|Minolta CS-100a-------|
-----------------|--------------|------------|--------------|-------------|--------------------------|
0.001-----------| 12,000----| 21,000----| 34,000-----| 55,000----|Photo Research--------|
-----------------| -------------| -----------| -------------| -------------|PR-705/715------------|
-----------------| -------------| -----------| -------------| -------------|(lab grade)-------------|
-----------------------------------------------------------------------------------------------------|

Just what does this show us? First, the ability of the measurement equipment to measure small amounts of light (Y2) drastically changes the numeric value of the ratio. Secondly, maximum light output also drastically changes the CR. Thirdly, even knowing what instrument was used and its capability, we have no reason to take any published value as an indication of display performance, as the above table shows a CR of 75 is the same as one of 12,000 with everything but the measuring device being equal. If, and only if, all displays were set up and measured identically with the same meter would there be any basis for an objective comparison as to their contrast ratio capability.

How does all this translate to the real world? Let's look at some contrast ratios that scientists have determined over several years of experimentation to be typical results. Since the scientists used lab grade equipment for their determinations, we'll use the same level of readings from our chart above. Film has a typical CR of 100:1 (most sensitive film is +- 120:1), video cameras produce results in the 30:1-40:1 range, film projectors can produce CRs of 200:1 or greater. Using these figures we see that any video projector (or other display technology) can easily reproduce the CR of our best source material (film).

How does the human eye compare to our source material in it's ability to discern the brightness dynamic range? The human eye on average sees life at a CR of 800:1, it can, when coming from a totally black environment to a bright one discern 1200:1 (this is called the chemical state CR). Obviously we can see a lot better than most of our instrumentation and certainly better than all of our video sources.

What does this mean in regards to the manufacturers contrast ratio claims?
Basically, you have no idea with what or how they made their measurements, whether the display was properly calibrated before doing the measuring, probably (or a random pick off the assembly line, doubtful). Therefore any device capable of achieving a CR of 120:1 or better will faithfully reproduce our current best source material when properly calibrated. When the display can produce a CR of 200:1 or greater then a faithful cinema experience can be enjoyed. Now should the device be able to achieve a ratio of 800:1 then you match what the eye sees in life. Anything else is over kill.

And over kill is exactly what we have in these claims! CRs of 500:1, 2,000:1, 3,000:1 and even 5,000:1 far exceed the human eyes capability to see, as well as that of our best film or theater. Practically every type of display we currently have is capable of showing us a lifelike dynamic range of light. What we need are better sources of video information, something capable of coming near the ability of our eyes, not display sources with inflated specifications.

Just remember to have your display system calibrated so that it will do the best possible job of displaying what is available.

Mel
DavLyn HTC

Joined:
Dec 4, 2002
Messages:
3
0
3. ### Jack Briggs Executive Producer

Joined:
Jun 3, 1999
Messages:
16,738
129
Well, Mel, let's see if your post can generate a thread here as opposed to referencing a discussion at AVS.

4. ### Michael TLV THX Video Instructor/Calibrator

Joined:
Mar 16, 2000
Messages:
2,909
1
Location:
Calgary, Alberta
Real Name:
Michael Chen
Greetings

CR specs are fairly useless in my mind.

If we all tend to agree that crt based displays have the best CR's ... and when we measure them the proper way, we end up with CR's in the range of 120 to 180 ...

For me, I then put so little weight on CR's specs for the other display technologies. A non issue.

Regards

5. ### Derek*k Auditioning

Joined:
Jul 15, 2003
Messages:
13
0
Well, on a PC, 256 shades of grey exsist from white to black. And beyond that, there are 256 shades (or intensities, I suppose) of each color. Does that not mean that the theoretical max for any 24bit image is 256:1?

I guess what I'm getting at is that 24-bit is regarded as "photorealistic" or "true color". Okay, fine. Then ought not any screen capable of 256:1 be deemed good enough?

I think where the big issue comes in is that often the contrast ratio is blamed for poor black level. 1200:1 from full on to full off is great, but not if it's referenced against something like (and please forgive my lack of proper terminology if need be), IRE 7 instead of IRE 0.

6. ### Mel W Auditioning

Joined:
Dec 4, 2002
Messages:
3
0
Let's see if I can this a little more clear than the original post. Digitization of a sample (256 shades) is not the same as measuring the difference between the lightest and darkest of a given color. That is the reason that VE, Avia and Home Theater Set Up and test gear also have what are called ramp signals. These signals are a smooth transition from dark to light if the digital TV/processor etc. has a poor circuit design then you see banding (digital sampling error, like the 256 shades). the 24 bit of photorealism is a bit misleading, not that is really inaccurate but rather people equate photos with what the eye sees. Two hugely different capabilities. One of the best examples to illustrate this is that CMYK printing and paint can contain all the colors that the eye sees, it is just a matter of mixing the right amounts of dye/pigment, Our good old Coca Cola can is printed at the factory but have you ever seen a photograph or TV image that even comes close to that color of red? You ask, well I've also not seen it in any printed ad either. True, but in advertising the products are photographed to make the ad. The act of taking a picture imposes a filtration guided by the limitations of the medium.

Now back to your main statement. Contrast ratio does seem to get blamed for poor black level. Conversly too many people use setting the black level as the cure-all for improving contrast ratio. Reality is not that simple however, the simple answer is that a poorly set black level reduces contrast ratio from the bottom end (crushing blacks)and an improperly set white level (contrast) reduces contrast ratio from the top end (crushing whites). Now take in to account the effect of viewing in a lighted room, this also reduces contrast ratio and the brighter the room is, the lower the ratio without changing a single setting on the TV.

As you can see, even simplified there are many things make this area murky and as we get better educated about contrast, and how to measure it, outside influences, the eye's instantaneous and total capability ratio it just gets worse. Take the table in my first posting and make the assumption that the FPTV can display a low level capable of being read by the lab grade meter and this is in a darkened room. Figure out what the ftL value is for that meter and then look at where the typical calibrators meter quits reading. The quality of the projector hasn't changed, the CR hasn't changed, but the reported CR has changed drastically. Since advertisers are notorious for making up numbers and even if they do get figures from their engineers how do we know what they were measured with and under what circumstances. That's why to me and I suspect others, published CRs are of no valid use what so ever. If we knew that the figures were derived under the same circumstances, by the same means and with the same instruments for all makes and models then we would have an objective number to make comparisons and buying decisions with.

7. ### Guy Kuo Supporting Actor

Joined:
Mar 6, 1999
Messages:
581
0
Thank you, Mel! That was spot on in my book.

I'll add a little bit about the 256 bit levels available, lest someone despair that we can only represent a ratio of 255:1 for the lowest to highest light level with 8 bits. Those 256 levels do not monotonically correspond to light intensities. Remember that there is also the gamma curve which is approximately a power of 2.3 to 2.4 exponential for a real CRT monitor. So the ratio between the lowest, non-black light level and the highest is approximated by....

Light output is proportional to i^ gamma

(1/255)^ 2.35 = 2.21e-06 or the lowest non-black
(255/255)^ 2.35 = 1 for max white

The ratio between the two levels of represented light is

1/2.21e-06 or 450,000 : 1

That's a pretty big dynamic range. Something which can be easily swamped out if the light level for "black" on a display is elevated. The non-linear mapping means 8 bits represent a considerably greater dyanmic range than if the mapping were linear. If the mapping was linear to light, the ratio between lowest and highest representable amount of light would be a scant 255:1

Don't get me wrong. More bits of accuracy would allow definitely allow finer representation of levels and smoother gradients. This is particularly important if signal processing includes gamma corrections. 8 bits isn't quite enough to do the job without errors during processing.