Histogram: Luminosity and Color

[MariaMaria]

Hello. I'm having some difficult fully understating how to interpret and analyze a luminosity and color histogram. I've read the tutorial but am still a little lost :-) Please help! Thanks in advance!

[Inkanyezi]

There isn't much to interpret in a histogram really.

As the histogram is only a statistical representation of pixel luminous levels, but does not say anything about where in the image they are, most of the histogram won't say much, and the histogram is completely useless if we don't also see the image. The image itself is the proof, the histogram only a help to see mostly one essential feature; whether too many pixels are burned in the image. The colour histogram will tell whether a saturated colour has lost tonality.

The only part of the histogram that can show this is the very last column at the far right.

If the last column of the histogram has any height at all, it indicates that the highest numerical representation has been hit somewhere in the image. If this column is high, it indicates that a larger number of pixels have reached this level. If many pixels are in this range, there is a probability that tonality is lost. It is not proof, because those pixels might be distributed over the image area and completely innocuous. When you look at the image itself, you can judge whether tonality is lost by the pixels that have reached the highest value.

Suspicion of lost tonality is higher when the histogram curve is rising toward the right end before the very last column, while a sloping histogram is less likely to indicate lost tonality, even if pixels reach the highest possible numerical level.

I mostly look at the red channel histogram when shooting flowers with very saturated colour, as I want to retain tonality and structure also in saturated areas. Also the blue channel sometimes will go high (i.e. rise at the far right), but I have not yet seen the green channel clipping.

A wee bit of interest can also be put to the far left side of the histogram. The first level is not significant, but those that come after. There's where shadow detail may be picked up. However, individual colour should be disregarded at the left end of the histogram. But again, it is when you try to adjust the curve and look at the image, that you will see whether shadow detail may be retrieved, and how much retrieval of shadow detail will influence the brighter parts of the image.

So in most respects the histogram is of little use, but it can help your judgement of the image sometimes, and it may indicate whether it is probable that you can retrieve tonality and detail.

[IzzieK]

Maria...I gave up on understanding the histogram a long time ago. So these days I just look at the blinkies on my image when I review it after the shot then adjust my EV till it is right. Or even the aperture or shutter speed, depends on what works for you...Not all photos have the same histogram as they vary in intensity of the darks and the light. Sometimes you get it right, sometimes you don't. The more reason why you should shoot RAW...tho on some occasion, extremes are unrecoverable you might as well change from high to low exposure to cover up your mistake if that wasn't the intended purpose of the shot...

My advice is very non-productive for you, but it does make sense to me...

[DanK]

I disagree. The histogram can be quite useful.

The bottom of the histogram represents tonality, ranging from pure black with no detail at the far left to pure white with no detail at the far right. The height of the spikes represents the number of pixels at a given level of brightness. (Technically, this is not quite right, but it is close enough.) For example, if your histogram shows a big mass in the middle, with nothing to either side, that tells you that you have a limited tonal range and that most of the pixels in the image are in that midrange. You have no pixels at the more extreme dark or light tonalities. You can use this information for several things:

--if the histogram hits the left edge, you have areas in the image that are pure black or "clipped", so you have lost detail.
--if the histogram hits the right edge, you have areas in the image that are "blown out" or "clipped"--pure white, so you have lost detail. The blinkies will also tell you that.
--if you have a histogram that hits both ends, you have more dynamic range in the scene than your camera can handle. You either sacrifice detail at one end or the other or combine images with HDR or exposure blending.
--if you have a histogram that does not approximately span the whole range, you have limited tonal range in the image. In that case, the location of the peak matters. In general, you are better off having the peak near the right than near the left ('expose to the right,' or ETTR) because the ratio of signal to noise is higher.

[Mike Buckley]

I've read the tutorial but am still a little lost :-)

You will probably receive more helpful responses if you indicate the particular issues that are confusing to you.

[benm]

I would like to add that on most cameras (there may be a few exceptions) the histogram is displaying the jpeg version of the RAW file. So even though the histogram is showing pixels hard up against the left or right side, indicating those pixels are "blown", it may be possible to recover them if the RAW file is custom processed.

[MariaMaria]

You will probably receive more helpful responses if you indicate the particular issues that are confusing to you.

I am having trouble understanding how the luminance histogram is calculated based on each color and how those colors relate to one another. The tutorial stated that each color is weighted using a different percentage; how does that translate into the photograph? How the RGB and Luminance histogram relate to one another, and also a general understanding of when to use each tool. Thanks!

[Mike Buckley]

The tutorial stated that each color is weighted using a difference percentage; how does that translate into the photograph?

That information may be helpful to know as a matter of interest to people who are technically-minded. (I'm not.) However, it's not necessary to know that information (I don't) when reviewing the histogram whether immediately after capturing an image or post-processing it.

How the RGB and Luminance histogram relate to one another

I think of the Luminance histogram as a composite (though I'm not sure that's entirely accurate) and the RGB histogram as a histogram of each channel. As an example, if the shadows or highlights are lost, the RGB histogram, unlike the Luminance histogram, will indicate which channels require your attention during post-processing.

a general understanding of when to use each tool.

I pay attention only to the RGB histogram because it's more informative, as explained above. I usually photograph still scenes (not action scenes), so I review the histogram after each capture so I know whether I need to change my exposure and retake the photo. I keep an eye on the histogram constantly as I'm post-processing so I don't accidentally lose detail in the shadows or highlights.

[MariaMaria]

Thank you for your reply! Very helpful. I am not as technically minded either I just want to make sure I'm making the most use of a tool.

[MariaMaria]

Thanks for your reply! So it seems like the RGB histogram would be a more useful tool generally speaking than the luminance histogram.

There isn't much to interpret in a histogram really.

As the histogram is only a statistical representation of pixel luminous levels, but does not say anything about where in the image they are, most of the histogram won't say much, and the histogram is completely useless if we don't also see the image. The image itself is the proof, the histogram only a help to see mostly one essential feature; whether too many pixels are burned in the image. The colour histogram will tell whether a saturated colour has lost tonality.

The only part of the histogram that can show this is the very last column at the far right.

If the last column of the histogram has any height at all, it indicates that the highest numerical representation has been hit somewhere in the image. If this column is high, it indicates that a larger number of pixels have reached this level. If many pixels are in this range, there is a probability that tonality is lost. It is not proof, because those pixels might be distributed over the image area and completely innocuous. When you look at the image itself, you can judge whether tonality is lost by the pixels that have reached the highest value.

Suspicion of lost tonality is higher when the histogram curve is rising toward the right end before the very last column, while a sloping histogram is less likely to indicate lost tonality, even if pixels reach the highest possible numerical level.

I mostly look at the red channel histogram when shooting flowers with very saturated colour, as I want to retain tonality and structure also in saturated areas. Also the blue channel sometimes will go high (i.e. rise at the far right), but I have not yet seen the green channel clipping.

A wee bit of interest can also be put to the far left side of the histogram. The first level is not significant, but those that come after. There's where shadow detail may be picked up. However, individual colour should be disregarded at the left end of the histogram. But again, it is when you try to adjust the curve and look at the image, that you will see whether shadow detail may be retrieved, and how much retrieval of shadow detail will influence the brighter parts of the image.

So in most respects the histogram is of little use, but it can help your judgement of the image sometimes, and it may indicate whether it is probable that you can retrieve tonality and detail.

[MariaMaria]

Very helpful!

--if the histogram hits the left edge, you have areas in the image that are pure black or "clipped", so you have lost detail.
--if the histogram hits the right edge, you have areas in the image that are "blown out" or "clipped"--pure white, so you have lost detail. The blinkies will also tell you that.
--if you have a histogram that hits both ends, you have more dynamic range in the scene than your camera can handle. You either sacrifice detail at one end or the other or combine images with HDR or exposure blending.
--if you have a histogram that does not approximately span the whole range, you have limited tonal range in the image. In that case, the location of the peak matters. In general, you are better off having the peak near the right than near the left ('expose to the right,' or ETTR) because the ratio of signal to noise is higher.

[Inkanyezi]

/snip/ In general, you are better off having the peak near the right than near the left ('expose to the right,' or ETTR) because the ratio of signal to noise is higher.

I snipped most of the text, but I think the mass of it should be disregarded.

As the histogram does not say a single thing about where in the image respective pixels are located, the histogram can have just about any form, but any "peak" might indicate low contrast in that luminosity region. A flat histogram might indicate high contrast.

But the worst is this total misunderstanding of ETTR, to expose to the right.

Exposing to the right is to expose the image so that only a very small fraction of the pixels will reach the right end, no matter how the rest of the histogram looks. An exposure done to the right will have a histogram that ends at the very right end, without reaching any height at all in the last column. Ideally one pixel should reach the numerical value 255.

It is sometimes counter-productive to expose to the right, but when the dynamic range of the sensor is smaller than that of the subject, exposure to the right will warrant that no highlight is blown. This is not always the best strategy. If the highlights that reach the highest numerical value are of no interest, if there is no detail there that you want to retain, but detail in darker parts of the image are more important, exposing to the right might be a bad idea. Better blow those highlights and get good rendition of what is important.

So exposing to the right is a method to utilise the largest possible dynamic range of the medium without blowing highlights. If the subject range is fourteen stops and your sensor range is ten stops, those last four will be completely black if you expose to the right. If you then let the highlights blow out about three stops, you will retrieve those same three in the dark parts of the image. Or you can take one image exposed to the right, with sooty shadows, and one four stops brighter with well rendered shadows. Combining the two to HDR and tone mapping can render a good image. Or you can get a Pentax K-5 or a Nikon D7000 or later in that series and you get fourteen stops in one exposure when you expose to the right and need only tone-map to get the same dynamic range.

So a high peak in the histogram may indicate low contrast in that particular luminosity. Higher contrast is possible when the histogram is flat, but it is not guaranteed, because the histogram does not in any way give a hint about where the respective pixels are situated. If they are randomly spread over the area, the image can have low contrast regardless of any peaks or being flat. A high contrast image always has a rather flat centre part of the histogram, but it can have peaks close to the ends. The histogram of a high contrast image can also be completely flat. A low contrast image mostly has one rather large hump somewhere in the middle of the histogram, sloping toward both ends.

So creative use of the histogram may be difficult to get to grips with, but it does help when you want to "expose to the right", and it will also hint to whether the dynamic range of the medium is sufficient for the subject range. It may also help you to see whether there is a lot of leeway to expose more.

[chauncey]

Andrew Rodney, AKA http://digitaldog.net/ represents the ultimate authority on color...click on the histogram tutorial. Warning, it takes a long time to load.

[Colin Southern]

Andrew Rodney, AKA http://digitaldog.net/ represents the ultimate authority on color...click on the histogram tutorial. Warning, it takes a long time to load.

I might add that in my opinion, he also has some "tunnel vision" on the subject; I've had arguments with him in the past, where he steadfastly maintains that processing in LAB colour offers no advantages whereas I maintain that it offers significant advantages. His position was also noted to differ from other "ultimate colour authorities" such as Bruce Fraser and Dan Margulis.

[Inkanyezi]

I think I forgot one thing, which Colin often points out.

When the dynamic range of the sensor is greater than that of the subject, there is little or no point trying to expose to the right. As long as the shadow values are about one stop or more above the threshold, it is better to have the histogram end a bit before the right wall than use it all the way to the end, as linearity of the three colours is not the same when you get into the region of the last stop. An eight stop dynamic range is contained within the dynamic range of any modern sensor at sufficiently low ISO, and it is enough (and could be better) to just keep it centered instead of trying to get it to the right.

[Dave Humphries]

Hi Maria,

The tutorial stated that each color is weighted using a difference percentage; how does that translate into the photograph?

The traditional values used in the television industry is that white is achieved by a ratio of 30% red, 59% green and 11% blue, however, you don't really need to know this, although it can occasionally be useful background information if thinking things through from first principles - for example it gives a clue why there are two green to every single red or blue sub-pixel in the bayer array used in most image sensors. I do not want to get into a debate over precise percentages or white balance issues, that's too deep for this thread (and for me to argue)
This is just an idea of the relationship between them, as requested.

So it seems like the RGB histogram would be a more useful tool generally speaking than the luminance histogram.

Many people, myself included, do find the RGB histogram more useful than the luminosity one.

Furthermore ...

In the real world - when the scene content is assessed alongside the (RGB) histogram, many of the aspects of an image can be helpfully judged as Dan suggests.

I snipped most of the text, but I think the mass of it should be disregarded.

As the histogram does not say a single thing about where in the image respective pixels are located, the histogram can have just about any form, but any "peak" might indicate low contrast in that luminosity region. A flat histogram might indicate high contrast.

The suggestion that just because, by assessing the histogram alone, you cannot say where pixels of a given value are, and this renders it useless as a tool, is applying an inappropriate level of scientific accuracy to the whole thing. No one was suggesting using ONLY the histogram, it is just one tool in our toolbox to help expose and subsequently process our images.

Sure - I can appreciate that a scene shot into the light, with sun sparkling off a myriad of wavelets means that the histogram is of less value than it is with a typical scene*, because it will show lots of clipping, but we as sentient beings can understand that those super bright pixels, being spread across the whole image area, are not telling us the whole story.

For example:





By "a typical scene", I mean where say, a white subject, like a swan, lit by light and areas that are over exposed do tend to be in one big clump and therefore often do matter.

For example, an image in which I have lost some detail where it matters:



There are other tools you may find helpful, for example, in spotting clipped highlights and shadows, most image editing software allows you to "false colour" indicate where on the image, pixels are 0 and 255, then make RAW adjustments to minimise these areas.

Many people find the histogram useful, so I'd "give it a whirl" (alongside "Mk.1 eye ball" on the image) and draw your own conclusions Maria.

I'm not going to mention ETTR here, since that's not really relevant to the original or subsequent questions from you and is covered in a separate tutorial, let's get one thing sorted before complicating things.

Hope that's helpful,

[MariaMaria]

Thank you for your detailed response and for the examples you've provided. I am not very tech savvy in regards to the more detailed workings of the camera's tools, pixel recording methods, etc. It just goes right over the top of my head I hope to learn more about those things at some point in the future but for now just need a general/practical understanding. Your responses help tie in some of the information from Dan, Mike, and others; very helpful to me, thanks!

[ajohnw]

In some ways the colour histograms are most important at the far right bright end. A lot depends one what the camera manufacture has done but if something is very bright green for instance clipping that channel wouldn't be a good idea. Some PP packages will not show single colour channel clipping in luminance mode. Camera might do the same thing or just display all pixel values irrespective of the colour they represent. Generally things aren't the particular red, green and blue that the camera happens to record but are some mix of these. If one channel clips the colours will shift. Going on cameras I have used luminosity would show clipping when this occurs.

The far left dark end is different. In a PP package trying to prevent any colour channel from clipping can result in very flat images. As the areas are very dark little colour information is needed so switching to a luminosity histogram that ignores single channel clipping makes sense. This is a known problem with people new to Rawtherapee that defaults to showing any channel clipping. I understand Adobe default to a mode similar to Rawtherapee's luminance mode in this region because I once adjusted an image done with Lightroom because it was clipping colour channels. In this case it was important that this was limited. Again though if the camera's luminosity histogram works as above it will show colour channels clipping at this end as well.

Shooting to the right via the histogram in many situations makes no sense at all. Say the camera gets the exposure correct, they often do in this situation, and the histogram shows space at each end - why brighten it in order to dim it down later. The method only works if there is bright white in the scene - or any other very bright colour. "Oddly" sites that point out how wonderful this method is always use scenes with significant white areas in them.

There is much to be said in favour of taking a shot and looking at the blinkies. If they are acceptable fine. If not adjust and shoot again. Often when some one goes to shoot some where at some time of the day one single adjustment will do for all shots taken. Past that where the nature of the subject needs specific adjustment nothing has changed.

LAB adjustments no good - the man must be mad. On the other hand I understand that PS's lab adjustments are nvg so maybe that is why he feels that way. Also some of them as they are usually presented can be very difficult to use effectively.

John
-

[DanK]

But the worst is this total misunderstanding

and you are being gratuitously disparaging because...?

As the histogram does not say a single thing about where in the image respective pixels are located, the histogram can have just about any form, but any "peak" might indicate low contrast in that luminosity region. A flat histogram might indicate high contrast.

A histogram is a form of a frequency distribution. It has nothing to do with the location of pixels, and neither I nor anyone else who posted said that it did.

Like any other graphical representation of a frequency distribution, it can't "take any form." The form is dictated by the frequency distribution. Other than one deliberate simplification, the histogram is what I said it was. If you create a graph on which the x axis has 256 values representing black to white, and you simply put one dot above each one of those values for every pixel that has that luminance, you would get a histogram that looks like the one on the camera, except that the one on the camera is of course scaled so that regardless of the distribution all bars fit within the vertical distance of the display.

But the worst is this total misunderstanding of ETTR, to expose to the right.

You say that ETTR "is to expose the image so that only a very small fraction of the pixels will reach the right end, no matter how the rest of the histogram looks." if that is how you use the phrase, that's fine, but that is not how I used it, and not what I wrote. I wrote: "In general, you are better off having the peak near the right than near the left." It is just a fact that the further to the right the histogram is, the higher the ratio of signal to noise. So, as long as you keep within sensible bounds, you are generally better off shifting the histogram to the right in terms of that one criterion (signal/noise ratio) because the large mass of pixels will have a good ratio. That is a valuable thing for the OP to learn, and it is a principle that I use all the time.

but any "peak" might indicate low contrast in that luminosity region. A flat histogram might indicate high contrast.

Keep in mind that the histogram is scaled, so the absolute height of the peak is not the real story. The story is the entire distribution shown by the histogram. My point was simply that if the histogram does not cover much of the range of the graph, the tonal range of the image as a whole is limited. If the x-axis were labeled, simple arithmetic would tell you how limited it is.

[Inkanyezi]

and you are being gratuitously disparaging because...?

Because what is quoted below is misleading. You mention a "peak" and you state that having that peak near the right is better than having it near the left.

--if you have a histogram that does not approximately span the whole range, you have limited tonal range in the image. In that case, the location of the peak matters. In general, you are better off having the peak near the right than near the left ('expose to the right,' or ETTR) because the ratio of signal to noise is higher.

Exposing to the right has nothing to do with any peak in the histogram, and there is no "best" position for a peak.

Arguably, having the whole histogram more to the right than to the left, when its values are completely contained within its limits, will supply more data in all levels of luminosity. Such a routine will decrease the risk of posterising when the image is edited.

The misleading part is the one about the peak.
It should be noted that I use the verb may. A peak may indicate low contrast. No peaks may indicate higher contrast. A high contrast image always has a rather löw and flat central part of the histogram and can have peaks nearer the ends of it. A high contrast image can have an almost flat histogram.
OTOH, a high hump, with both flanks sloping down toward the ends, is typical for low contrast images. Nothing of that will tell us much that we won't readily see in the image and already know, but it tells us something about the histogram and that we cannot judge much about the image from it. It should be evident that the shape of the histogram is no determinant of image quality. There is no such thing as a "good" or "bad" shape of a histogram, it only is.

So what we are looking for is that the entire histogram at best is contained within it borders, in order to have all tonal values of the subject rendered in the image.

However it is not mandatory that all subject tones should be rendered. Often we won't bother about shadow detail. Shadows may be rendered pitch black, and this can be used for graphical impact. The same goes for highlight. Sometimes highlight detail is not important. We create images, and there are no "recipes" for a good image; our whole art is visual.

/.../ I can appreciate that a scene shot into the light, with sun sparkling off a myriad of wavelets means that the histogram is of less value than it is with a typical scene*, because it will show lots of clipping, but we as sentient beings can understand that those super bright pixels, being spread across the whole image area, are not telling us the whole story.

For example:


/.../

The histogram in the above high contrast image is low and rather flat over the whole central part, just as I stated above, and it has two peaks, at the very ends. The peak to the right indicates that there are many pixels that are clipped, and the peak to the left tells us that there are many pixels that are pitch black. We have already observed visually that many pixels are clipped, and we can also suspect that there are pitch black areas, but none of that is important information. The impact of the image is what we see when looking at it.

I said before, that a peak in the histogram may indicate low contrast at that luminosity value. This is also confirmed by Dave's image. The myriad of sharp highlights do not have any contrast at all within them. They are simply clipped and completely white.

There is no particular position to be preferred of a peak in the histogram. The peak mostly should be disregarded.

Exposing to the right is the technique to utilise as much as possible of the available sensitivity of the medium to the right side of the histogram, without clipping.

It should be evident, that when we allow highlights to clip substantially, we are not using that technique.

And still, any peak in the histogram is of no interest, and it is not "better" to have it to the right. The peak should simply be where it belongs. So if the image has lots of dark values, the peak is probably at the left side and should be there.