Histogram ... a basic question

[Abitconfused]

If a histogram displays luminance values then isn't it incorrect to say that a histrgram has a black point because a black point has no luminance value to display. It also seems that a histogram does not display a white point because the liuminance of a white point could be confused with clipping. Or is a white point clipped by definition?

[Colin Southern]

If a histogram displays luminance values then isn't it incorrect to say that a histogram has a black point because a black point has no luminance value to display.

It's probably a moot point because "technically" is one thing but "real world" is another, from a couple of angles (1) Just because there are areas of the image with values above zero doesn't mean that we'll be able to discern detail in those (low) areas, and (2) Just because an image has some pure black areas doesn't mean to say they'll be there in sufficient numbers for the eye to register it as having sufficient contrast. As such, when setting the black point of an image - on a correctly profiled monitor - in my opinion, it's better to just do it visually (but using the histogram as a guide if one wishes).

It also seems that a histogram does not display a white point because the luminance of a white point could be confused with clipping. Or is a white point clipped by definition?

White point / clipping is same thing. In essence both black and white points both simply define areas of zero detail.

[xpatUSA]

If a histogram displays luminance values then isn't it incorrect to say that a histogram has a black point because a black point has no luminance value to display. It also seems that a histogram does not display a white point because the liuminance of a white point could be confused with clipping. Or is a white point clipped by definition?

The OP terminology is bit loose; a pedant would likely claim that a "white point" is the bit in the middle of a color space that has no hue, and that "black point" is something to do with TV screens. In reality, when the terms "black point" and "white point" are used, they tend to mean something other than what a histogram displays. And therein could lie a bit of confusion for people reading the post.

Lets say you take a shot in the "real world" of a gray cat sitting in an ash-pile. If you can't find that scene, let's say you shoot a close-up of a Kodak gray card. With reasonable exposure, the luminance histogram should show a peaky distribution roughly centered in the horizontal brightness axis. You can't really say that either end of that distribution is a "point", black or white. However, you could say that the "white point of a histogram" is the far right column which has a value of 255 and the "black point of a histogram" vice-versa with a value of 0, but the statement doesn't tell you much, 'cuz you already know what those column values are.

So I'm disagreeing by saying that, if a histogram can be said to have a "black point", it is the far left column which has a value of zero and you can't get blacker than that!

And, at the other end, the "white point" simply has a value of 255 - but the histogram itself has no way of knowing whether the cause of that value is clipping or not. That is to say, a histogram's white point is not "clipped by definition" - it is just white.

[Bryan Conner]

The OP terminology is bit loose; ... That is to say, a histogram's white point is not "clipped by definition" - it is just white.

Ted is correct.

A histogram is only a graph showing the relative levels count (vertical height) over a range from 0-255. It only shows you what is there, it is up to the viewer to interpret the relevance of the information. A strictly luminance histogram does not tell you much information, only that there is black or white in the image. A color (RGB) histogram can tell you if a single channel is clipped. Pure white (255,255,255) can not be clipped.

[Colin Southern]

Pure white (255,255,255) can not be clipped.

In theory.

In practice it will almost certainly be a "place holder" for data that would have sat at "256 or beyond" if the scale went past 255 (ie anything at 255 probably represents loss of detail in a real-world photo).

[Bryan Conner]

In theory.

In practice it will almost certainly be a "place holder" for data that would have sat at "256 or beyond" if the scale went past 255 (ie anything at 255 probably represents loss of detail in a real-world photo).

This is interesting. How can there be detail even at 255,255,255? Of course, I am speaking in the realm of Photoshop. Photoshop only uses RGB values from 0,0,0 to 255,255,255. 256 is not possible with Photoshop.

[Colin Southern]

This is interesting. How can there be detail even at 255,255,255? Of course, I am speaking in the realm of Photoshop. Photoshop only uses RGB values from 0,0,0 to 255,255,255. 256 is not possible with Photoshop.

When the pixels at 255 are in amongst pixels of a lower value; it simply represent the highest possible value of a range of values.

By "256" I was just trying to articulate that anything at 255 probably "should have been" 256 - 280 - 294 - 221 etc (ie the original scene had detail there, but it was lost due to clipping) or "anything on the histogram at 255 probably represents clipped data (unless it represents a white border or something similar). The chances of capturing a scene (or even manipulating) a scene so a highlight area "just happens" to be EXACTLY 255 is pretty remote.

[xpatUSA]

By "256" I was just trying to articulate that anything at 255 probably "should have been" 256 - 280 - 294 - 221 etc (ie the original scene had detail there, but it was lost due to clipping) or "anything on the histogram at 255 probably represents clipped data (unless it represents a white border or something similar). The chances of capturing a scene (or even manipulating) a scene so a highlight area "just happens" to be EXACTLY 255 is pretty remote.

In spite of Colin's bias toward the "real world" I have to agree because the probabilities associated with a distribution are indeed represented by a histogram.

I was going to post a black-and-white image of an Ace of Spades as a rebuttal. Imagine my irritation when not one real-world shot of such a card had a histogram consisting of only 0's and 255's!!

A moment's thought says "of course not". Even if I shot a perfectly homogeneous white border and got the exposure exactly right, the output (not clipped) from the sensor would show as a normal distribution on an image's histogram, albeit with a very small standard deviation (i.e. it would look peaky). This would be because the random nature of photon conversion, deviations of pixel performance, the addition of noise, etc., ad naus.

Let's say the real "white" came out in the image centered around a column-value of 245 on the histogram. With a standard deviation of 2, there would be very few pixels - if any - at the 255 level. Most of the pixels would be in range 239 - 251 i.e. +/- 3 standard deviations. Indeed, it would be an accurate image to the best possible extent and still not clipped, but to some folks it might appear "not white enough". As soon as the hand reaches for the mouse in order to shove the histogram to the right, clipping will almost certainly occur.

Thus we see that the shape of the histogram is important when assessing an image's distribution. Which, in turn, means that the presence of 0's or 255's per se can not actually be regarded in isolation. For example, a bathtub shape is likely clipped to hell at both ends, because probability says so. On the other hand, an upside-down bathtub shape, while perhaps not the best of images, has quite a low probability of clipping.

[Colin Southern]

In spite of Colin's bias toward the "real world"

The alternative to "real world photography" is photography only of test targets so that cameras and lenses can be dissected to the nth degree - and the resulting images pixel-peeped at 400% before the "results" are written up in the form of a "review" to help sell magazines and promote websites.

My personal opinion is that this "measurbation" does far more damage than it does good as it leads folks to believe that "the numbers" are important parameters when it comes to image quality (and who doesn't want the highest quality images?), when in reality, the lab rats concentrate on the minor variables and completely ignore major variables like photographer knowledge & skill and quality lighting.

When's the last time a head to head review of a Nikon D8000 and Canon 5D3 concluded with a summary of "they're both great cameras - it doesn't matter which one you choose - and if you REALLY want the best quality images, go buy a bunch of lighting equipment and go sign up at Kelby Training". Just ain't gunna happen is it - why? Because the objective of the non-real-world reviewers is never to help the photographer ...

... only to sell the magazine or promote the website.

I think I'll stick to real-world photography - producing real-world images - for real-world money and the real-world enjoyment of myself & others

[xpatUSA]

My personal opinion is that this "measurbation" [!] does far more damage than it does good as it leads folks to believe that "the numbers" are important parameters when it comes to image quality (and who doesn't want the highest quality images?), when in reality, the lab rats [!] concentrate on the minor variables [!] and completely ignore [!] major variables like photographer knowledge & skill and quality lighting.

[ . . . .]

I think I'll stick to real-world photography - producing real-world images - for real-world money and the real-world enjoyment of myself & others

Bit of an over-reaction that, don't you think, especially as the rest of the post was actually agreeing with you.

If it is claimed that the sole arbiter of photographic image quality is human vision, un-aided by any analytic means, then what I've read about the performance of the eye in distinguishing detail and color has me quite worried.

Sorta like flying by the seat of your pants blindfolded :-)

[Colin Southern]

Bit of an over-reaction that, don't you think, especially as the rest of the post was actually agreeing with you.

Not at all in my opinion Ted. I'm not "aiming it at you" - just making the point that in my opinion that when it comes to photography people spend far too much time worrying about inconsequential things whilst completely ignoring the major variables.

If it is claimed that the sole arbiter of photographic image quality is human vision, un-aided by any analytic means, then what I've read about the performance of the eye in distinguishing detail and color has me quite worried.

Sorta like flying by the seat of your pants blindfolded :-)

Well perhaps we need to ask "why do we engage in photography"? Is it to create soul-satisfying images that lift our mood - make us think - spur emotions? Or do we do it to get closer to some perceived mathematical idea?

[xpatUSA]

Not at all in my opinion Ted. I'm not "aiming it at you"

I'm pleased to hear that Colin because, at my age, self-gratification ain't what it used to be.

just making the point that in my opinion that when it comes to photography people spend far too much time worrying about inconsequential things whilst completely ignoring the major variables.

An extreme view which perhaps aims to balance the opinions of those "lab rats" who, apparently, never take pictures with any merit due to their ignorance of "the major variables".

Well perhaps we need to ask "why do we engage in photography"? Is it to create soul-satisfying images that lift our mood - make us think - spur emotions? Or do we do it to get closer to some perceived mathematical idea?

Nice spin at the tail end

Perhaps we could instead ask "why are we interested in how cameras work, in the physics of light and the human perception thereof, in the minutia of color spaces, and all the technical stuff?" Is it so we [geeks] can show off on fora such as this? Or, do we do it so we can take better pictures?

[Colin Southern]

An extreme view which perhaps aims to balance the opinions of those "lab rats" who, apparently, never take pictures with any merit due to their ignorance of "the major variables".

My position (and those of most other professional photographers too I might add) has always been that photographer skill plays a far more significant role in determining the quality of a photo than does minor variances in equipment specifications. If that's an "extreme view" in your opinion then I respect that opinion, but may I respectfully suggest that you probably won't get a lot of support for that line of thinking.

I'm quite sure the lab rat boffins CAN take photos of merit - but - that's not what they're doing nor saying in their reviews; all that they ever seem to promulgate are meaningless examples of 100% crops to show the difference between "X" and "Y" ... leaving the reader with the impression that "model A" is superior to "model B" because of some insane metric that can never be seen in a real-world photo. What they NEVER say is "it doesn't matter a damn that camera A has a 5.7 micron pixel pitch compared to camera B's 5.65 - so toss a coin to decide between them - but if you REALLY want to produce better images then don't waste your money buying our crappy magazine - sign up for some good photography training instead". The first approach sells magazines & generates clicks ... the second doesn't. It's not "dishonest" per se, but in my opinion it's not ethical either. And if anyone isn't convinced then for proof they need look no further than their nearest book store - pick up just about any photo magazine and they'll find it crammed full of paid advertisements - host/beneficiary "competitions" - technically accurate but practically worthless reviews - and usually some plain vanilla "photography tips".

Perhaps we could instead ask "why are we interested in how cameras work, in the physics of light and the human perception thereof, in the minutia of color spaces, and all the technical stuff?" Is it so we [geeks] can show off on fora such as this? Or, do we do it so we can take better pictures?

The latter of course, but I'm really not sure why you've brought that up or what relevance it has to the current conversation.

[ajohnw]

As I understand it black and white points are terms usually used in respect to the tonal range extracted from a raw file.

White and black point are also used in relation to monitors and have the same meaning in a way as they do with raw files only they are the limitations of the set up and the white point would usually have a colour temperature associated with it.

Applied to a histogram a black point would be a thin black line full height on the extreme left of the histogram that in simple terms can't be over exposed. The white point would be the converse but could be over exposed and would look just the same. Maybe they should make them flash if they are. Only problem though is I suspect they don't really know until the exposure is actually taken so can only predict that it will be.

Equal rgb values for pure white probably don't exist deep within cameras or monitors as colour temperature is largely taken care of via variations in blue levels / relative changes to r and g.

-

[Mike Buckley]

Colin,

Your views about this stuff are refreshing from my point of view. I subscribe to two magazines that have been providing the sort of content that you describe for decades and I'm sure for the reasons that you explain. I skip past the objectionable content as I have always done (it's easily recognizable at a glance) and proceed immediately to the helpful, enjoyable content.

[Colin Southern]

Colin,

Your views about this stuff are refreshing from my point of view. I subscribe to two magazines that have been providing the sort of content that you describe for decades and I'm sure for the reasons that you explain. I skip past the objectionable content as I have always done (it's easily recognizable at a glance) and proceed immediately to the helpful, enjoyable content.

Thanks Mike,

To be honest, I've given up buying them now -- it's like being stuck in "ground hog day". I seem to get more faster and more targeted content from selected web sources these days.

[xpatUSA]

My position (and those of most other professional photographers too I might add) has always been that photographer skill plays a far more significant role in determining the quality of a photo than does minor variances in equipment specifications. If that's an "extreme view" in your opinion then I respect that opinion, but may I respectfully suggest that you probably won't get a lot of support for that line of thinking.

Hello Colin, I do respect your position in spite of the occasional dig. I too am well aware that there are excellent shots taken with technically inferior equipment and poor shots taken with the best that money can buy. When two equipments have only "minor variances" then the more skilled photographer wins, quite obviously. Especially if a "minor variance" is a 0.05 um difference in pixel pitch, i.e. less than 1% difference.

An interesting condemnation of 100% crops - a possible reference to dpReview's test scene and their 100% crops of the wine bottle label? Ignoring the relevance of a wine bottle to the "real world", surely those of us who shoot products rather than natural scenes, models, etc. would find dpReview's crops relevant and of interest?

Well perhaps we need to ask "why do we engage in photography"? Is it to create soul-satisfying images that lift our mood - make us think - spur emotions? Or do we do it to get closer to some perceived mathematical idea?

Perhaps we could instead ask "why are we interested in how cameras work, in the physics of light and the human perception thereof, in the minutia of color spaces, and all the technical stuff?" Is it so we [geeks] can show off on fora such as this? Or, do we do it so we can take better pictures?

The latter of course, but I'm really not sure why you've brought that up or what relevance it has to the current conversation.

English sarcasm at it's finest, sorry. A simple paraphrasing of your earlier point, of no real value to the discussion.

[groovesection]

Applied to a histogram a black point would be a thin black line full height on the extreme left of the histogram that in simple terms can't be over exposed. The white point would be the converse but could be over exposed and would look just the same. Maybe they should make them flash if they are. Only problem though is I suspect they don't really know until the exposure is actually taken so can only predict that it will be.
-

I am sure you know, But on Canon cameras if you bring up the image, then hit Disp twice it will bring up the histogram and also the "blinkies"
The blinkies will show you (by flashing black) that an area in the image is over exposed.
So then just a case of dropping the EV a couple of stops to compensate

[xpatUSA]

Couldn't find a gray cat but I was cutting and burning wood yesterday, so I took an ash-pile shot . . .

. . . carefully avoiding this critter, of course!



Do forgive the shaky cell-phone shot and less-than-professional composition, my hand was about a foot away from a venomous Southern Copperhead.

So here's the ash-pile:



Just a snap with the old D50 + kit lens, no IQ critique, please. Opened in ACR 5.4, all sliders zero except WB.

There we see a classically-shaped Gaussian distribution and just the beginnings of clipping at the right of the distribution, indicated by a taller column at the end of an otherwise gradual decline. The taller column represents a discontinuity in the nicely-ordered world of statistics, telling us that all is not well with the image (as if that were not obvious from the red bits in the preview). If, instead, the curve had smoothly fallen to a small pixel count at the 255 column the picture would have been better from an exposure point of view.

Of somewhat lesser interest is the large width of the distribution, considering that the scene has very little apparent contrast.

[Colin Southern]

An interesting condemnation of 100% crops - a possible reference to dpReview's test scene and their 100% crops of the wine bottle label? Ignoring the relevance of a wine bottle to the "real world", surely those of us who shoot products rather than natural scenes, models, etc. would find dpReview's crops relevant and of interest?

Hi Ted,

No - I'm not referencing DPReviews tests (I don't visit their site particularly often) - just referring to how the majority of reviews (especially lens reviews) rely on 100% crops to highlight things like edge sharpness or chromatic abberation etc. The irony is that they HAVE to use 100% crops because the performance (or lack of it) that they're trying to highlight just wouldn't be visible if the full-resolution image were supplied but only viewed at "normal" internet sizes (not defined, but can we agree that most online images may be in the order of around 800 - 1200 pixels), nor in a normal sized print (perhaps something between 6 x 4 inches and 8 x 12 inches?). In other words they need to present the image in an "abnormal" way to illustrate their point but the irony is that in the real world images are never/seldom used in this way. If they were forced to present "the whole picture" then they may as well not bother because they'd be forced to concede that it doesn't make any "real world" difference.

For the information to be visible in "the real world" one would need a monitor with a resolution that matched the camera's sensor and was big enough for our eyes to be able to resolve (perhaps 5 or 6 feet wide) or a print that was approaching the same dimensions. And to my mind, that's just not relevant to every-day photography - including product photography. It's a bit like comparing two vacuum cleaners by looking at sections of cleaned carpet under the microscope when all the buyer wants to be able to do is stand at the door and have it look clean. In the case of photography, far too many people "try to save the pixels and in the process ruin the image" (eg shooting at too low an ISO to avoid the "noise" and in the process getting camera shake or DoF issues, or buying lens "B" over lens "A" because it's "sharper" and then not sharpening the image correctly in post-production because they haven't invested in appropriate post-production skills (with sharpening having 10 times more impact on the quality of the finished image than any inherent difference between any two lenses) - or purchasing an expensive camera because reviews show that it's "superior", but not investing in appropriate lighting equipment or how to use it (when investing in the lighting equipment and how to use it would have produced a 100 times improvement in the image compared to the difference between an expensive camera and a far cheaper camera. As Lance Armstrong said "it's not about the bike" (although as it turns out it might be more correct to say "it was about the drugs", but I digress ...).

By the way, product photography is a big part of my business (more so than portraiture than landscape) - and "real world" is still very much the order of the day where (for example) the inherent sharpness (as identified by 100% crops of lens "A" compared to lens "B" pales into (literal) insignificance compared to the differences in image quality due to compromises of things like camera to subject distance (further away to improve DoF, but closer to improve resolution), and of course lighting. I WILL often work at magnifications approximating 100% - but only to clean up an image so that it's guaranteed to look good at lower magnifications.

In closing, here's a point for folks to ponder ...

What do you think would be the effect on the world's photography if it were impossible to zoom (or view) an image beyond, say, 25%? (and 100% crops didn't exist) Folks wouldn't be able to compare things like CA and edge sharpness). Would it make any difference to the photography of the worlds top shooters? I don't think it would make the slightest bit of difference. I do think it would force people to concentrate on the BIG PICTURE first and foremost though, and I think that that could only be a good thing. Too many photographers "major in minor things", and I'm really not to sure why; it's almost as if they're extremely capable of listing all of the factors that potentially affect quality, but (perhaps due to a lack of experience) seem totally incapable of giving appropriate weighting to those factors so things like "photosite noise influenced by pixel pitch as mentioned in the review" gets more weight than "correct lighting". If folks really want to improve their photography than - again, my opinion only - I think they should stop inspecting their images at insane magnifications - stop trying to choose equipment based on trivial specifications - and start concentrating on the "big picture".

Just my opinion anyway.