Westminster fountain at sunset

by Jorge Córdoba

submit your photo


Hall of Fame
View past winners from this year

Please participate in Meta
and help us grow.

Take the 2-minute tour ×
Photography Stack Exchange is a question and answer site for professional, enthusiast and amateur photographers. It's 100% free, no registration required.

For having more contrast in an image, it's good to have a wide range (i.e. difference between highest and lowest) of light intensities in a picture; and for keeping details, neither highlights nor shadows should be clipped.

But are there any other, more specific, guidelines on how distribution of light levels would affect perception of a photograph? Are some histogram shapes more visually appealing than others? E.g. when setting up lighting, should I strive for a flat histogram, perfectly symmetrical histogram (a normal distribution), or local peaks at both ends?

share|improve this question
    
The 1st assertion in the 1st paragraph my be not completely true, although for everyday situations it makes much sense. Think about a histogram where there are two narrow peaks - one at the left edge and one at the right (w/o clipping). The image is very contrasty but the range of intensities is small. –  ysap Jun 13 '11 at 11:14
    
@ysap yes, I did mean "range" in its statistical meaning (difference between lowest and highest value), not in the mathematical sense (set of input values) - histogram is a tool from statistics. Edited, perhaps it's more clear now. –  Imre Jun 13 '11 at 12:46
    
@Imre - further thinking here - I think the criterion should be large variance and not a wide range, as you can imagine an image where the intensities are spread allover, but mostly are concentrated in a narrow range. –  ysap Jun 13 '11 at 16:02
3  
Somewhat related, slightly off topic, but very interesting: ironicsans.com/2007/10/reader_takes_histogram_idea_ev.html –  Andres Jun 13 '11 at 18:40
    
@Andres - for the last hour I am wandering the web (one link leading to another), following your link. Some very cool stuff and a great idea for a project! –  ysap Jun 13 '11 at 21:38

4 Answers 4

up vote 12 down vote accepted

In short, there's no useful connection. The histogram shows a certain view of the information in an image, and it's useful for avoiding some specific problems and can be used in image analysis, but without referring to the original image, you can't really tell if a particular histogram shape is good or bad (or even if things which look like they might be problems are actually problems).

The look of the histogram will depend on your desired result, and there are many possible desired results which are perfectly valid. A high-key photograph will be shifted towards the right side of the image, with almost nothing on the left (which represents the shadows). A low-key photograph will be the opposite. As that link goes into excruciating detail over, this is not to be confused with high-key lighting, which will generally produce a quite even histogram; or with low-key lighting, which will produce a largely left-side histogram with some spikes on the right. As you can easily imagine, all of these might produce aesthetically-appealing results, and there's no particular right answer (even for the same subject).

The histogram is a tool for visualizing the information in an image "sliced" in a certain way — it discards certain information (in this case, spatial/location information) in order to make other information more clear. But that information is actually required for making the kind of aesthetic judgment you're looking for. The histogram shows you the relative amounts of darkness and brightness, but you really need to see where the light falls to decide if an image works or not.

A visualization tool that might be more useful for this sort of thing is a grayscale "pixelated" diagram of the image (simply made by converting to grayscale, scaling down to a tiny number of pixels, and then back up again).

Guen with BalloonGray Pixelated Diagram

It's pretty obvious that there's still no right answer here for one pixel diagram to be better than another, but I think it can be a useful way for thinking about the placement of light in the image. The example I chose is one where the histogram looks basically fine, but the resulting image doesn't quite live up to what I was aiming for. The grid visualization is another way of thinking about the overall lighting that I do find helpful. (Do I really want that splotch of brightness in the bottom right?) It's not perfect, of course — while the dark bar in the right-middle might draw me to look at whether that shadow is problematic (and I think it is), it glosses over other also-important lighting details, like the shadow of my daughter's legs on the log, for example.

Tracing the image (either programatically or by hand) to produce a block diagram with actual shapes of bright areas and shadows might be even better. One particular problem I notice here is that the subject's face is split between "pixels", making that bright spot less apparent. If I did the diagram by hand, that'd a clear shape. Of course, that's a lot more work!

Here's the histogram for the same image:

Balloon image histogram

This shows that the overall exposure is pretty good, with detail retained in the shadows — some blown-out highlights, but not many. That happens to be the specular reflections of the lights on the balloons — but there's no way of telling that from the histogram. And there's another weird bright spike, which turns out to be where the red and green channels are blown out in the yellow flames. Also, you can see that there's some other bumps — the one towards the right turns out to be mostly the result of the large blue balloon. If the balloon were a different shade, that "hill" in the histogram would be located elsewhere, making the shape of the overall histogram different — but the overall image would not be very different at all.

To me, the difference between the kind of analysis you can do with the histogram and the analysis you can do in looking at the lighting and composition using a spacial representation of some sort shows the answer to your question. The histogram is interesting and useful, but the view it gives isn't such that you can make decisions about the perception of the image as a whole from it alone. You can see if there are certain problems, and you can see if the image as a whole is high-key or low-key, but beyond that there's just not the right data to tell if a certain histogram distribution corresponds to a a lighting situation that works or to one that doesn't.

On the histogram, as you say, keeping the edges from clipping is important, because you can't do anything with values that aren't there. And having an even histogram to start gives you more data to work with. In addition to those things, expose-to-the-right advocates point out that the linear nature of digital sensors means you'll get more room to work if you weight your histogram more to the right side (still without clipping). (But that's advice for working with images intended to be processed into images with "normal" histograms as output, so I don't think it really applies to what you're asking.)

share|improve this answer
    
it may be that I am too tired now, but I could not see how this answers @Imre's question. If the answer is there, you may want to make the point a little bit clearer. Assuming we know what is a histogram and how to read it, I believe Imre's intention is to ask what characteristics of an histogram should we try to have in order to get a specific effect in our image. Specific effect here not necessarily means greatest contrast or such. I believe the question is in general - a mapping from histogram shape to a typical output. –  ysap Jun 13 '11 at 2:05
    
@ysap: there isn't one, is the point. I'll try and look for a few examples of very different images with similar histograms. –  mattdm Jun 13 '11 at 2:11
1  
Agree with @mattdm here: there is no relationship between histogram patterns and the 'appeal' of a photo. This is like saying that the gas chormatograph pattern of a given food impacts its taste. There isn't a direct relationship, as they are both measurements of the same data: one is the output of your brain's evaluation of all the pararmeters that make up the image (recognition, color, clarity, emotion etc) vs the computer's evaluation of the balance of dark and light pixels. So the image is driving the results of both measurements, but there is no relation between the two outcomes. –  cmason Jun 13 '11 at 15:39
    
"So the image is driving the results of both measurements, but there is no relation between the two outcomes." sums it up well — thanks. –  mattdm Jun 13 '11 at 15:49

There's a maximum contrast ratio a sensor can record. It's usually better to minimise contrast with lighting and boost it in post if necessary.

The best histogram shape depends on the subject, again when I'm shooting a studio setup with lights I aim for a flat histogram to maximize the options for post processing.

share|improve this answer

For having more contrast in an image, it's good to have a wide range of light intensities in a picture; and for keeping details, neither highlights nor shadows should be clipped.

That's about it. You cannot calculate what's aesthetically pleasing, thus I recommend looking at the picture instead of the histogram.

share|improve this answer

In her article Top 10 Painting Tips for Strong Compositions tip #3 Marion Boddy-Evans suggests to divide the scene into areas of light, dark, and mid-tones. For a strong composition, total of areas in each category should be significantly different, exact order being unimportant.

In terms of histogram it means that when it's divided into three parts along horizontal axis, each part should have a proportion clearly different from other parts.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.