What does the "group" think of the attached illustration?

[Abitconfused]

Is this illustration correct?

[Digital]

I have never seen a histogram look like the theoretical bell shaped curve. Is this the representation of the perfect picture as far as pixel distribution goes?


Bruce

[Abitconfused]

I suppose if you averaged a number of histograms according to observed tones, a curve of some kind would reveal itself. Since most photograph's are not of a black dog in a coal mine or a snowman against a white fence, middle tones should dominate...at least in theory. Why else calibrate a light meter to middle gray? Such a decision would frustrate us if most of us went out at midnight to do our photography. So, in my thinking, a theoretical framework is called for if just for the sake of discussion.

[William W]

Is this illustration correct?

The illustration is neither correct nor incorrect.
Without any context, the illustration is nonsense.
We need to know what is the ‘theoretical photograph’ and what is the purpose that the theoretical nature serves.

What is the source of the illustration?
What is the context of the illustration, within that source?

WW

[Abitconfused]

How can a middle gray be illustrated if no illustration illustrates a middle gray?

[Abitconfused]

This photograph illustrates the curve. Or does the curve illustrate the photograph?

[pnodrog]

Mid point grey (18% or 14% depending on your belief system..) is the exposure point at which you have the best chance of keeping the tonal range of a "normal scene" within the dynamic range of the film or sensor. It has no direct relationship to the distribution of tones within the scene but just relates to the extents of the tonal range.

At a guess if you analysed the average tonal range of photographs taken by the general public the the histogram would probably be skewed towards the lighter tones due to the predominant inclusion of sky and Caucasian skin tones. Certainly if you did a survey based on photographs taken by mountaineers and skiers (provided they managed to correct the exposure properly) it would show a significant bias of light tones. Photographs taken of the moon and stars would have a completely opposite tonal distribution.

[AB26]

Ed,

The drawing is an illustration of the distribution of tones in a theoretical photograph. In theory it could be an ideal histogram.
Though the ideal histogram might not be attainable in real life situations the illustration is a good indication of what a histogram tells you.

If looked at in the right context and understood as such, the illustration might be helpful in better understanding dynamic range and how to use a histogram to get “correct” exposure.

[pnodrog]

This photograph illustrates the curve. Or does the curve illustrate the photograph?

I am surprised at the tonal range in this example photograph. The sky looks darker than it should be in relation to the shadow under the tree. Have you used shadows/highlights or done some other adjustments in pp?

[Abitconfused]

Good observation. This was modified in the latest version of Photo Ninja. I was impressed by the ability of the program to deliver greater tonal depth in highlights without skewing the entire image adversely (IMHO). It presents a sort of semi HDR that may or may not suit individual taste or complement the desires of the photographer.

[pnodrog]

Good observation. This was modified in the latest version of Photo Ninja. I was impressed by the ability of the program to deliver greater tonal depth in highlights without skewing the entire image adversely (IMHO). It presents a sort of semi HDR that may or may not suit individual taste or complement the desires of the photographer.

I think the software has done a reasonably good job but it is not representative of the tonal range in a "normal scene" ...

[Abitconfused]

I think you are right and I think the histogram reveals the "defect." The larger number of pixels in the highlight and white point areas betray the alteration. Thus the difficulty in generalizing an "ideal" histogram. No such creature can exist. Neither can a "perfect" histogram. But a histogram for the sake of discussion can exist and that is what is offered initially.

[Manfred M]

A histogram shows the luminance data of an image as it is. It in no way shows what an "ideal" image looks like.

As others have said, this diagram is quite meaningless unless there is some context around it. If some is suggesting that this wit the way a "perfect" scene would come out like, then again, more nonsense.

[pnodrog]

I think you are right and I think the histogram reveals the "defect." The larger number of pixels in the highlight and white point areas betray the alteration. Thus the difficulty in generalizing an "ideal" histogram. No such creature can exist. Neither can a "perfect" histogram. But a histogram for the sake of discussion can exist and that is what is offered initially.

Some of the best understanding and learning is achieved by creating a model or theory(actually I should say hypothesis) and then seeing if it works and if not then working out what is wrong with it. It is actually the basis of nearly all scientific advancement....

[Saorsa]

I think that there is an error by the originator of the 'histogram' because they chose to use a 'normal' curve to illustrate their point. Is used to illustrate some concept of normalcy and is hardly ever achieved except in very, very large datasets.

A histogram is simply a count of some range of values. It is useful at the boundaries but the values in the middle are just counts with some utility.

This image would not show a lot of data on the right side.



and this one would not show a lot on the left



If the author wanted to discuss histograms they should have used a real histogram from an image and discussed it in terms of value counts.

[DanK]

It is not a sensible illustration.

Many things in nature are roughly normally distributed. This isn't one of them. The histogram will show the frequency distribution of pixels, arranged by luminance. There is no reason to expect that to be normal, as the illustrations show.

[Abitconfused]

I bet if you distributed the luminance of 100,000 photographs it would reveal a curve very much like the normal curve. For everyone who takes a photograph outdoors (tending to push the histogram to the right) there is someone who attempts to take a photograph indoors without a flash and receives an underexposed image (tending to push the histogram to the left). You see this at sport stadiums all the time. Flashes go off far from the photographer's subject of interest and an underexposed image is the result. Of course the above normal curve reveals abnormality (hideously underexposed or overexposed images) as much as it does normal (if we describe normal as an acceptably exposed image). This does not even begin in the least to describe any individual image as good or bad or nice or not nice or moral or immoral or ideal or not ideal. It is simply a distribution of luminance. Similarly, a normal curve displays intelligence with the center having an IQ of 100...sad. Fortunately, photographers have a much higher IQ as revealed in the above comments.

[AB26]

The above graphical illustration in a real life image.

[DanK]

Andre,

That is because your 'image" is random noise, roughly centered on a neutral gray. Some random processes generate a normal distribution. Most people's photographs, however, aren't random information, although sometimes I think some of mine come disturbingly close.

Dan

The above graphical illustration in a real life image.

[DanK]

I bet if you distributed the luminance of 100,000 photographs it would reveal a curve very much like the normal curve.

It depends on how you aggregated. If you took the means of thousands of histograms and then calculated the distribution of those means, it seems to me that this should be approximately normal, although not necessarily with a mean (that is, the mean of the distributions of means) at the center. That's the central limit theorem: with sufficient numbers, the distribution of the means of distributions approaches normality even if the parent distributions are non-normal. However, if you just added the histograms (to be more precise, if you added the frequencies underlying each of the histograms), my guess is that you would end up with something much more like a uniform distribution. Not that this is something I will test.