What constitutes a 'correct' histogram?

[AlwaysOnAuto]

I shot some more quilt pictures this morning for my wife. It was overcast out, no shadows. I took the pictures with my D7000 using my Micro-Nikkor 55 2.8 lens. Most shots were @f8-11.
When I look at them in my rudimentary photo software program, the histograms all seem to be biased towards the left end of the graph. When I change the exposure +2 or so, the histogram is more spread out across the graph, the image looks 'brighter'.
Am I supposed to try and the histogram spread out on the graph?
Here's a sample shot that I 'fixed'.

[Manfred M]

Alan - the histogram tells you a lot of things, but there is no "right shape" to one. The histogram is what it is.

By the way, there are two types of histograms, there is a luminance histogram (I assume this is what you are looking at) as well as a set of histograms that show the distribution of each of the three colour channels.

When you take a picture, ideally, you would like to have a histogram that has no data (or more precisely minimal data) at the extreme left (black) or extreme right (white) sides. If your histogram bangs against the left hand side, you are clipping (i.e. have lost data). The same is true on the right hand side.

A histogram that has has a lot of light tones (as in your image), I would expect to see it biased to the right hand side. If it predominantly has dark tones, I would expect it to be biased towards the left hand side.

[dje]

Am I supposed to try and the histogram spread out on the graph?
Here's a sample shot that I 'fixed'.

Not specifically Alan. I suggest you adjust the exposure slider (if necessary) until the image looks about right in terms of "brightness". If you are increasing the exposure in post, just make sure you don't cause white clipping by checking the right hand side of the histogram for spikes (at 255).

Dave

[dennybeall]

The Histogram is a simple counting of picture elements. The right side is pure white elements and the left side is pure black. Your original histogram was rightly weighted to the right because of all the white in the picture. When you forced some of the white to get darker, as seen in the photo you presented, the histogram gets more even BUT the white part of the photo get grey and ripples show. So, if nothing is crawling up the ends you're good.

[Mike Buckley]

The correct histogram of a documentary image that as much as reasonably possible replicates the scene being photographed will display the distribution of tones on the graph that is consistent with the tones in the scene being photographed; if the displayed distribution of tones in the histogram is inconsistent with the scene and if the image is primarily documentary, the histogram is incorrect and can be made correct by using a different exposure.

I strongly recommend reviewing the CiC tutorial about the use of the histogram.

[Steaphany]

As long as the histogram doesn't climb the wall at either end, you are doing fine. When you do see darks piling up on the left, or brights piling up on the right, that indicates under or over exposure and an unrecoverable loss of image detail.

If the range doesn't span the whole histogram, then the exposure is limited to a lower range of contrast.

Sounds like the adjustment you performed expanded the dynamic range of the image and since there were darker darks, your brights project a sense of being comparatively brighter.

[davidedric]

A couple of replies talk about the original image being biased towards the right, but the OP says it was towards the left.

Is this to do with the white areas being interpreted as grey, or something else?

Dave

[Tringa]

I believe you are correct Dave.

A large amount of the subject is white or light in tone and the camera's meter detected a bright scene and adjusted the exposure because it "assumes" all scenes reflect 18% of the light falling on them.

It is the same effect that occurs when taking a snow scene. Without some positive exposure compensation the snow comes out looking a bit grey rather than white.

Alan, I'm guessing you are photographing the quilt with the possibility of selling it so it is a matter of setting an exposure that reproduces the colours as correctly as possible.

As already noted above, there is no correct histogram. It depends on the subject. If the subject has not light tones then the right side of the histogram is going to be largely empty and vice versa.

Dave

[chauncey]

FWIW...when I shoot white birds, my RGB numbers are held to 248.

[DanK]

Alan,

I agree with Mike--the best starting point is this site's tutorial on histograms. However, I will throw in a few points:

You need to distinguish between the shape of the histogram, including how spread out it is, and its location in the graph space.

The histogram is essentially, but not quite, a display of the count of pixels at each of the 256 values of an 8-bit image (2^8=256). I say "not quite" because the camera is doing a few things other than counting. First, your camera is translating the raw image into the 8-bit jpeg represented by the histogram. Second, your camera is probably rescaling the Y axis to avoid or at least lessen the number of bars that hit the top of the display.

Assuming that the histogram doesn't hit either end, it's width is determined by the tonal range of the image. An image that is mostly at one level of brightness will be narrow, with a high peak (hence the need for rescaling), while an image with a wide range of brightness will have a histogram that is spread out. That is as it should be. The histogram is simply representing the range of illumination in the image. You may or may not want to leave it alone. For example, images with very narrow histograms tend to be very drab and low-contrast, as Steaphany pointed out. It is easy to fix this in postprocessing. You can stretch the histogram to give a wider tonal range, and you can use tools like the curves and contrast slide to change where in the image the transition from light to dark is concentrated.

Now, assume you have an image with a narrow histogram. It's location in the graph space will be determined by exposure. Increase exposure by any means, including adding a positive EC, and the histogram will move to the right. In general, as several people have said, you will have more data in your image relative to the amount of noise if you keep the histogram as far right as you can without hitting the end. (To be precise, some people will keep it a bit lower, say, maxing out at 245 or 250, but let's leave that aside for now.)

If your histogram hits the end, you will "clip," which is the photographic term for what is called "censoring" in math. Let's say that you hit the bottom. As your exposure gets darker, more and more pixels that would be spread out among different values if you had a better exposure will be piled up together at 0, because there is nowhere lower to go. You lose all that information. The same happens if you clip at the high end.

Assuming that your histogram is narrow, changing exposure would in theory simply shift the histogram right or left without changing its shape. That won't happen. It will also change the shape, but usually only a bit. That's because your histogram isn't displaying what the camera captures, but rather the distribution of luminance in the jpeg it has created by using whatever postprocessing algorithm (picture style, or whatever) you have set the camera to use.

So you should set the exposure to position the histogram in a good place--no clipping at either end, and fairly close to the top. You should use postprocessing to change the spread, and other aspects of the shape of the distribution, to taste. Assuming you don't clip, you can't do much in camera to change the spread, other than changing your selection of postprocessing algorithm in the camera, which will have a slight effect. That's just one of many reasons that most folks who post here rely on doing the processing themselves.

Dan

[Dave Humphries]

Alan,

Mostly good advice above, only one person seems to have mis-understood what you were saying, but even their last comment is good.


I think we should consider why it happened though ...

There's a lot of white in that image, so (if left to its own devices), the camera meter will have guided you to expose (in Manual mode) that to reproduce a mid grey (this is how it works) - not what you want (for this scene) and that explains the original histograms being biassed to the left.

Adding +2 EC* in camera, or increasing it in PP, will bring the whites up to roughly where they should be, as seen in the image above and as the 'now spread further to the right' histogram demonstrates.
* only relevant if shooting in a semi-auto mode; P, A or S (or Auto-ISO in M).

This is exactly the situation where, until you have the experience to predict the result of shooting a scene like this, you should be reviewing the shot immediately after capture using the camera's histogram and adjusting exposure upwards for a second capture at the correct exposure, using the guidance provided/linked above.

HTH, Dave

PS
The only caveat to the general advice on how to use luminance histograms is when shooting scenes with a predominant bold colour (esp. red), e.g. a flower in sunlight taking up most of the frame, the luminance histogram will be insufficient to guide you and that's when you must use the RGB histogram and ensure that none of the individual colours are clipping (esp. on the right hand side).

[Mike Buckley]

You should use postprocessing to change the spread, and other aspects of the shape of the distribution, to taste.

To clarify my take on that, once I am post-processing the image, I never try to change the shape of the histogram as an end goal. Instead, I try to make the image look the way I want it to appear, not caring about the resulting shape of the histogram. However, during post-processing I will always keep an eye on the histogram but only as a visual guide that helps me ensure that I am maintaining the black and white points (the far left and right sides of the histogram, respectively) as I want them for that particular image.

until you have the experience to predict the result of shooting a scene like this, you should be reviewing the shot immediately after capture using the camera's histogram and adjusting exposure upwards for a second capture at the correct exposure

Despite that I consider myself to have extensive experience and that I usually predict the histogram correctly within about one-half stop, I am blown away about how wrong my prediction can sometimes be, so much so that once I do achieve the desired histogram I am completely stymied as to why the exposure had to be changed so dramatically. So, when shooting a scene that includes no action, I always review the histogram immediately after and adjust the exposure accordingly if needed. When shooting a scene with action by holding down the shutter release for multiple bursts, I always review the histograms of that group and adjust the exposure if needed before capturing the next group of images.

[DanK]

To clarify my take on that, once I am post-processing the image, I never try to change the shape of the histogram as an end goal. Instead, I try to make the image look the way I want it to appear, not caring about the resulting shape of the histogram

Good catch. I agree. I didn't mean to imply otherwise. The key is 'to taste'. E.g., if I look at an image and decide that it lacks impact because it lacks tonal range, I then adjust that, the effect of which is to change the shape of the histogram.

[AlwaysOnAuto]

Thanks for the replies all.
I had gone thru the tutorial here on histograms, but that was a while ago.
The shot I posted shows my set up, and yes, there is a lot of white in it. I tried to get the quilt to fill the view as much as possible but still have a white 'border' around it so my wife can crop it more easily. These are shots for her scrap book of all her quilts. I should have taken a number of shots of all of them but was trying to beat the sun coming out/burning off the overcast as shooting in direct sun would have been much harder I think. As it is, playing with the exposure slider seems to be working as far as getting the colors to show as bright as they really are.
Thanks again for all the comments.

[Mike Buckley]

That scrapbook must be absolutely fascinating!

[AlwaysOnAuto]

It is being worked on right now on our family room carpet.



It is a quite elaborate piece of work itself, let alone the pictures of the quilts in it.

[Urbanflyer]

Several folks, more experienced than I, have suggested "no clipping" as a part of reading the histogram. One instructor whose work I admire had a slightly different take. He suggested that clipping was only a problem if it compromised the image being produced. Some slight clipping might be acceptable if areas where no detail of importance was to be found if the net result was an image which worked for the photographer; i.e. clipping in a minor area of deep shadow. I have found this works for me though I do not apply his approach without thought.

[DanK]

Several folks, more experienced than I, have suggested "no clipping" as a part of reading the histogram. One instructor whose work I admire had a slightly different take. He suggested that clipping was only a problem if it compromised the image being produced. Some slight clipping might be acceptable if areas where no detail of importance was to be found if the net result was an image which worked for the photographer; i.e. clipping in a minor area of deep shadow. I have found this works for me though I do not apply his approach without thought.

I agree. The same can be true of clipping specular highlights. The key is understanding that the price is that there is no data when you clip.

[dan marchant]

Several folks, more experienced than I, have suggested "no clipping" as a part of reading the histogram. One instructor whose work I admire had a slightly different take. He suggested that clipping was only a problem if it compromised the image being produced. Some slight clipping might be acceptable .....

Massive amounts of clipping are acceptable, if that is what the photographer wants. High key and low key images often result in massive amounts of clipping and information being "lost".

[Manfred M]

Massive amounts of clipping are acceptable, if that is what the photographer wants. High key and low key images often result in massive amounts of clipping and information being "lost".

I would have to STRONGLY disagree with this statement. A high key image is a properly exposed image that predominantly has light colors shot against a light coloured background. A low key image is a properly exposed image that predominantly has dark colors shot against a dark background. What you are describing is incorrect. Loss of detail at that scale is quite unacceptable in any image.

Massive overexposure is just that and does not result in a high key image. Massive underexposure does not result in a low key image.