How can achieve better low-light shots?

[Mike Buckley]

Sergio - it takes the right tools and techniques to become a good low light photographer.

To Manfred's point, it occurs to me that you're trying to master one of the more difficult types of photography, probably unknowingly so, before first mastering the basics. I recommend that you gain a clear understanding of how to use a histogram to your advantage, of the minimum shutter speed that will allow you to handhold your lens on its own or when stabilizing the camera on a support, and of how to check your aperture setting to ensure that you've got the largest possible aperture that also keeps everything you want to be sharp in focus using regular daylight. Once you've done that (though probably not before), you'll find photographing low-light scenes much easier, enjoyable and with greater success.

[Shadowman]

To Manfred's point, it occurs to me that you're trying to master one of the more difficult types of photography, probably unknowingly so, before first mastering the basics. I recommend that you gain a clear understanding of how to use a histogram to your advantage, of the minimum shutter speed that will allow you to handhold your lens on its own or when stabilizing the camera on a support, and of how to check your aperture setting to ensure that you've got the largest possible aperture that also keeps everything you want to be sharp in focus using regular daylight. Once you've done that (though probably not before), you'll find photographing low-light scenes much easier, enjoyable and with greater success.

To me the lowlight histogram is more difficult to interpret, trying for a perfectly exposed lowlight shot usually results in a flooded light look or animated capture, even with an image with a good splash of lighting (thirty percent of the composition or more) still might result in a histogram that leans to the left. CiC has a good tutorial on the histogram and a good example of lowlight histograms.

[Manfred M]

To me the lowlight histogram is more difficult to interpret, trying for a perfectly exposed lowlight shot usually results in a flooded light look or animated capture, even with an image with a good splash of lighting (thirty percent of the composition or more) still might result in a histogram that leans to the left. CiC has a good tutorial on the histogram and a good example of lowlight histograms.

Honestly, the only thing I really watch for in a histogram is the loss of shadow detail or highlight clipping in each of the three colour channels. I also prefer (for technical reasons) to have the luminosity histogram biased to the right, although in low light work, this is often not the case as the tones tend to be darker and are more concentrated on the left hand side of the histogram. If there are bright lights in a low light shot, there is a good chance that there will be some highlight clipping.

That more or less sums up how I use histograms in low light work as well as in regular work during image capture. I take a slightly different approach when I do my editing.

[xpatUSA]

To me the lowlight histogram is more difficult to interpret, trying for a perfectly exposed lowlight shot usually results in a flooded light look or animated capture, even with an image with a good splash of lighting (thirty percent of the composition or more) still might result in a histogram that leans to the left. CiC has a good tutorial on the histogram and a good example of lowlight histograms.

Sorry, John, I am confused. What is a "low light histogram"? Is it like in RawDigger where you can select a range of levels for display, or does it just mean a histogram of a photo taken in low light?

[Shadowman]

Sorry, John, I am confused. What is a "low light histogram"? Is it like in RawDigger where you can select a range of levels for display, or does it just mean a histogram of a photo taken in low light?

Mike,

It's a histogram taken in low light. Taken in A or S mode the histogram will tend to lean to the left unless I get a warning, if I shoot in manual exposure mode and I try for a bell curve the final output in my opinion doesn't look like the actual scene and at times appears animated in its capture. If I expose for the lights, I can sometimes prevent clipping but the output may not make a favorable looking print or display.

[Manfred M]

I share Teds bewilderment over the concept of a "low light histogram". A histogram is just a graph showing the distribution of data from dark on the left to light on the right. There is no such thing as a "correct" histogram.

Taken in A or S mode the histogram will tend to lean to the left unless I get a warning,

Have you tried using exposure compensation when shooting in A or S (or P) modes? If it is dark out, the camera's light meter will tend to overexpose, so try applying -1 to -3 exposure compensation to move the histogram to the left. There is no need to shoot manually as this is what EC is there for.

if I shoot in manual exposure mode and I try for a bell curve.

Why? Unless you have a very unusual scene, a bell curve shape is the last thing I would expect. Here is a recent night shot where the histogram shows a lot of data towards the dark side with a tiny bit of loss of shadow detail and a tiny spike on the light end and some reasonable level of mid-tones (largely thanks to the snow acting as a reflector). Get rid of the snow and the mid-tones will have less of a presence, but certainly nothing that looks even close to a bell curve.

[DanK]

It's a histogram taken in low light. Taken in A or S mode the histogram will tend to lean to the left

I think you should separate the shape of the histogram from its location (left to right). An image that has a lot of its pixels dark relative to other pixels in that image will have a peak at the left and right skew, like Manfred's example. This has nothing to do with how light or dark the scene is. It's just a function of the relative luminance of pixels. [An exception to this: if you over or underexpose to the point of clipping, you will change the shape of the histogram because pixels pile up at the extreme.] You can them move that histogram right or left, changing its location, by changing the exposure. And of course in postprocessing, you can change the shape.

In night photography, I find that it is best to expose to the right, bringing the right tail of the histogram close to the maximum. this will give the highest ratio of signal to noise. If the image looks to bright overall, one can just darken it in post, without doing any damage to image quality.

[Shadowman]

I share Teds bewilderment over the concept of a "low light histogram". A histogram is just a graph showing the distribution of data from dark on the left to light on the right. There is no such thing as a "correct" histogram.



Have you tried using exposure compensation when shooting in A or S (or P) modes? If it is dark out, the camera's light meter will tend to overexpose, so try applying -1 to -3 exposure compensation to move the histogram to the left. There is no need to shoot manually as this is what EC is there for.



Why? Unless you have a very unusual scene, a bell curve shape is the last thing I would expect. Here is a recent night shot where the histogram shows a lot of data towards the dark side with a tiny bit of loss of shadow detail and a tiny spike on the light end and some reasonable level of mid-tones (largely thanks to the snow acting as a reflector). Get rid of the snow and the mid-tones will have less of a presence, but certainly nothing that looks even close to a bell curve.

Hi Manfred,

Yes I have used EC to compensate. When I spoke of a bell curved histogram I meant in daylight, I wouldn't expect that same shape in lowlight and this goes back to instructing the OP to learn how to interpret a histogram; in lowlight the interpretation totally changes. I was perhaps talking about two different concepts in one thread.

[Manfred M]

Hi Manfred,

Yes I have used EC to compensate. When I spoke of a bell curved histogram I meant in daylight, I wouldn't expect that same shape in lowlight and this goes back to instructing the OP to learn how to interpret a histogram; in lowlight the interpretation totally changes. I was perhaps talking about two different concepts in one thread.

Okay - here's a SOOC raw file taken in daylight conditions. No bell curve there either.



Daylight does not give you a bell curve either. The only way you can get a bell curve image is the image has a lot of mid tones and no dark or light tones, with a a drop off in both the highlights and shadows. I know I have seen bell shaped curves in text books, but I really can't recall ever having seen one in an image that I've taken.

[Mike Buckley]

I imagine that a lot of people without an engineering background probably find histograms more confusing than helpful.

In the event the OP doesn't have an engineering background, I mention the following only so the OP understands that no such thing is required: If you can understand a graph comprised of an X-axis and a Y-axis, you can understand a histogram. That's because that is all a histogram is. Engineering backgrounds are not required to understand that, as is evidenced by me who has no technical background and is instead blessed mostly by music and sales backgrounds.

[Shadowman]

I've come close quite a few times, it's not a prerequisite just a warm fuzzy when I do. An example of your comment on lots of mid tones, I don't think I ever bothered processing as on the day of this shoot my friend and I were looking for prime colors on an overcast day. We processed everything with vibrant reds and yellows couldn't find any vibrant blues that day.

[DanK]

I've come close quite a few times, it's not a prerequisite just a warm fuzzy when I do.

No reason why it should give you a warm fuzzy. There is nothing inherently better about a bell-curve-shaped histogram. it's only better if that is the distribution of luminance that the particular image calls for. This shape does not in itself indicate a good exposure; it's the location of the histogram on the x-axis that does.

Manfred's posts show concrete examples of this. One can have well-exposed images with any shape of histogram--unimodal, bimodal, trimodal, symmetrical, skewed, or whatever.

There are two parts to this. The first is the histogram of the capture. The most important thing is that it not clip at either end, but if there is room--if it doesn't extend to both extremes--it is usually better to have it toward the right-hand edge, that is, ETTR. One controls this, but not the shape of the distribution, with exposure.

The second part is the output histogram, the distribution of luminance you want in the final image. That can be very different in shape. For example, the typical S-curve adjustment changes the shape of the histogram, compressing the tails and expanding the middle. (I wish software would show the output histogram on the Y-axis; this would make it much clearer. You can change it in all sorts of ways.

This matters for Serge's original post. It's common in night photography to have the mass at the left--because a lot of the image is dark sky. In mathematical terms, the histogram will often be right-skewed. This is nothing to worry about; it's what the shape should be, if much of the sky is dark. The reason to look at the histogram is to figure out whether exposure is optimal, and to see that, you have to look at how far it extends to the left and right. That's what Serge should worry about.

[Manfred M]

At the risk of offending, I believe that these statements are misleading, Manfred, sorry. One only has to consider a bright but misty scene with say one small dark object (park bench for example) and the high levels (the mist) will almost certainly have a normal distribution - i.e. the so-called bell curve. Many histograms with large areas, each homogenous but at different levels will show several normal distributions in the same histogram. For example noisy shots of the color checker card might do it. Easy to with my cameras.

Perhaps the term "bell curve" is being used here in the restricted sense that it covers the whole range of levels - as opposed to shadow noise for example, or shot noise in the sky?

I imagine a lot of people without an engineering background find histograms more confusing than helpful.

Ted - I think I might not be clear enough in what I have written and I'm coming at it from a slightly different perspective than you are, I suspect. I absolutely 100% agree that a histogram can indeed have a shape of a normal distribution. The issue that I am trying to address is that I've run into quite a number of people who seem to think that the histogram (I assume they are referring to the luminance histogram) should have a "bell curve shape" and they can't understand why their image doesn't show that shape.

The answer is obvious, but that requires an understanding as to what the histogram display is actually showing rather than what people think it should look like.

[Manfred M]

I've come close quite a few times, it's not a prerequisite just a warm fuzzy when I do. An example of your comment on lots of mid tones, I don't think I ever bothered processing as on the day of this shoot my friend and I were looking for prime colors on an overcast day. We processed everything with vibrant reds and yellows couldn't find any vibrant blues that day.

+1 to Dan's comment, especially the two first sentences.

You are looking at a luminance histogram here and that is a calculated value of 0.30R + 0.59G + 0.11B. There is absolutely nothing special about it other than the calculation that looks at the three colour channels just happen to end up looking this way.

[xpatUSA]

+1 to Dan's comment, especially the two first sentences.

You are looking at a luminance histogram here and that is a calculated value of 0.39R + 0.59G + 0.11B. There is absolutely nothing special about it other than the calculation that looks at the three colour channels just happen to end up looking this way.

The weighting coefficient for R seems a little high, a typo perhaps?

For JPEGs, Y' = 0.299R+0.587G+0.114B where the coefficients should add up to 1.

[Manfred M]

The weighting coefficient for R seems a little high, a typo perhaps?

For JPEGs, Y' = 0.299R+0.587G+0.114B where the coefficients should add up to 1.

Yes a typo - the fingers are sometimes faster than the brain. The "9" and "0" key are right beside each other.

[xpatUSA]

In the event the OP doesn't have an engineering background, I mention the following only so the OP understands that no such thing is required: If you can understand a graph comprised of an X-axis and a Y-axis, you can understand a histogram. That's because that is all a histogram is. Engineering backgrounds are not required to understand that, as is evidenced by me who has no technical background and is instead blessed mostly by music and sales backgrounds.

Sorry that I have apparently implied that an Engineering Degree is required in order to understand a histogram.

Indeed anyone can do with a little study. I have withdrawn the comment.

[DanK]

I just started a new thread with photos and histograms that I hope makes some of this clearer.

[Mike Buckley]

Dan's new thread can be viewed by clicking this link.

[DanK]

Dan's new thread can be viewed by clicking this link.

Thanks, Mike. I should have thought to do that.