Histograms

[Jase]

Can someone please help me understand how to read and work with a histogram properly? If you can explain it with a few images that would be highly appreciated.

Thanks in advance!

 

[Ted Chu]

Histograms--Good Explanation

Histograms--Good Explanation

http://www.luminous-landscape.com/tutorials/understanding-series/understanding-histograms.shtml

 

[Luis Ortiz]

This link may help you.

www.valleyviewfinders.org/powerpoint/histogram_sly.ppt

 

[Paul Leeming]

Expose the image such that the right side of the histogram gets as close to the edge of the graph as possible but NOT over. This method is commonly used in digital photography and goes by the acronym ETTR (expose to the right).

Simplified explanation:

You do this because the nature of digital sensors is that each stop down from full white uses half the values of the previous stop to store its colour information, limited by its bit depth (eg Red = 12bit sensor so in theory 12 stops are possible, though this is never reached due to other factors). So using 2^12 you can have 4096 possible colour/luma values. The top stop uses 2048 of these to describe its colour and thus is the most tonally accurate. The next stop down uses 1024, the one after that 512 and so on. Each stop down introduces less precision and more chance of an incorrect colour being noted, resulting in what we know as noise.

If you underexpose Red by 1 stop you have effectively halved your dynamic range as the top 2048 values are sitting idle.

Use the histogram in conjunction with the False Color Meter to see WHERE your exposure is in the image. That way you can make the decision to overexpose a part of the shot (such as a light) knowing that your subject exposure is still maximised.

All this results in what looks to the layman's eye as a very dull, flat image when you first open it up in your editing program. But in fact that's exactly what you want! The flat image has the fatest dynamic range and appropriate use of curves can quickly bring it into gorgeous relief while retaining all the subtlety of its colour spectrum:

HTH!

Paul

 

[Rob van Gelder]

Paul, this is so true, if only I could convince our Thai clients /dop's to think like this . But for some reason they do not seem to grasp the logic of reduced dynamic range when under exposing and that they have to "sort of forget" about those extreme highlights like lamps in shot when making decisions about exposure.....

so after a shooting they come back, complaining the image is noisy....

 

[Jase]

Those links and explanations really helped. Thanks!

 

[Von Thomas]

Wow Paul!!

This was the best written explanation of exposing for digital I have come across.

I'm constantly trying to make my clients understand that they lose color fidelity/accuracy when they stop down too much trying avoid blowing the highlights. Blowing highlights is bad, but the lack of that missing data, should you stop down too much, can cause destruction to your image.

 

[Dan Hudgins]

Can it fit?

Can it fit?

But for some reason they do not seem to grasp the logic of reduced dynamic range when under exposing and that they have to "sort of forget" about those extreme highlights like lamps in shot when making decisions about exposure.....

so after a shooting they come back, complaining the image is noisy....

I got a sample image made with another camera that had two lamps in the shot both off the top of the curve and a dark interior through a doorway in the center of the frame with a window into the night.

That frame falls off both ends of the Histogram, so just moving the Histogram left or right will not help.

What would help would be to reduce the wattage of the bulbs in the lamps, or gel them, and to light the interior properly.

Even if you have a camera with 16bits, that is not a substitute for good cinematic lighting inside or out.

I asked before what advantage there was to shooting film when you are going to scan it anyway, and the answer was that the film makes the Histogram narrower so "everything" fits, and that is true for our scanner from normal negatives, or I can make the scanners Digital Camera's exposure bracketing wider for high contrast negatives getting more out of the highlights and shadows than would be possable with normal Film to Film printing.

With only 12bits in a Digital Cinema camera you cannot always underexpose enough to hold all the highlights in high dynamic range natural scenes, so the Histogram may be wider than the sensors range, in that case, I think looking at the image on a monitor would help, as would maybe using more reflectors or fill lighting so you can stop down a little without getting too much noise in the dark areas...

 

[Von Thomas]

It's true with still as it is in motion, you need a properly lit scene. If your lamps are causing your histogram to clip on the right edge, ND those puppies, and add more light to the interior, to bring the shadow detail up.

Neg film has more latitude, you can get away with things you can't shooting chrome or digital. That said, a properly exposed scene, be it film or digital, is a beautiful thing. Light it right, expose it right, you have no problems in post.

 

[Casper Warnich]

Expose the image such that the right side of the histogram gets as close to the edge of the graph as possible but NOT over. This method is commonly used in digital photography and goes by the acronym ETTR (expose to the right).

Simplified explanation:

You do this because the nature of digital sensors is that each stop down from full white uses half the values of the previous stop to store its colour information, limited by its bit depth (eg Red = 12bit sensor so in theory 12 stops are possible, though this is never reached due to other factors). So using 2^12 you can have 4096 possible colour/luma values. The top stop uses 2048 of these to describe its colour and thus is the most tonally accurate. The next stop down uses 1024, the one after that 512 and so on. Each stop down introduces less precision and more chance of an incorrect colour being noted, resulting in what we know as noise.

If you underexpose Red by 1 stop you have effectively halved your dynamic range as the top 2048 values are sitting idle.

Use the histogram in conjunction with the False Color Meter to see WHERE your exposure is in the image. That way you can make the decision to overexpose a part of the shot (such as a light) knowing that your subject exposure is still maximised.

All this results in what looks to the layman's eye as a very dull, flat image when you first open it up in your editing program. But in fact that's exactly what you want! The flat image has the fatest dynamic range and appropriate use of curves can quickly bring it into gorgeous relief while retaining all the subtlety of its colour spectrum:

HTH!

Paul

Great post. I learned a lot.

I did however google it, and this guy thinks it's a myth:
http://www.rags-int-inc.com/PhotoTechStuff/ETTR/

But he doesn't really adress your statement that the highest stop contains the most information.

I'm getting all confused :umm:

 

[Dan Hudgins]

Binary...

Binary...

But he doesn't really adress your statement that the highest stop contains the most information.

If you have ever watched a Binary counter you will see that small numbers advance the high lamp faster than large numbers. If you start from 0 and count up the time it takes for the next higher lamp to light gets longer and longer.

So in 12bit data half the counts are between the two highest bits, the two lowest bits have no counts between them.

This leads to "posterization" in the dark areas and too much information in the bright areas, too much since the eye is less and less able to resolve small changes in bright areas.

So digital sensors that are linear, like the one used in RED ONE, have too little tone value in the shadow areas and too much in the bright areas.

To fix that you need a log pre-amp before the A2D converter and a lower noise sensor element.

There is no drawback to having "extra" data in the bright areas, you just round off the top of the curve in post to lighten the mid-tones. Film and your eyes are not linear, nor are monitors etc. so you do not want linear values from the sensor A2D in the end result as viewed most of the time...

 

[Daniel Browning]

Thanks for the post, Paul. I agree that ETTR is a good idea, but I think you've cited the wrong reasons why.

You do this because the nature of digital sensors is that each stop down from full white uses half the values of the previous stop to store its colour information, limited by its bit depth (eg Red = 12bit sensor so in theory 12 stops are possible, though this is never reached due to other factors). So using 2^12 you can have 4096 possible colour/luma values. The top stop uses 2048 of these to describe its colour and thus is the most tonally accurate. The next stop down uses 1024, the one after that 512 and so on. Each stop down introduces less precision and more chance of an incorrect colour being noted, resulting in what we know as noise.

The kind of noise you are describing is quantization noise. No well-designed raw camera, including RED, exhibits this type of noise. There are two other types of noise, called photon shot noise and read noise, which are so much greater in power that quantization noise never has the chance to rear its head. If read noise were improved so much that quantization noise could ever become visible, then RED (or whatever manufacturer) would have to upgrade their ADC to a higher bit depth. But they know to do that already. The read noise in the RED ONE is such as that 12 bits accurately encode all the data available. More bits would *not* help.

So the reason to ETTR is to reduce the other types of noise, which have nothing to do with the number of levels and everything to do with the amount of light hitting the sensor. More light = less noise.

For example, you can take a RED ONE image with +1 exposure, truncate it to 10 bits (still linear), and it will have less noise than a 12-bit image with +0 exposure.

For further reading I highly recommend these two threads:

http://www.reduser.net/forum/showthread.php?t=23568


http://reduser.net/forum/showthread.php?t=24088

 

[David Rasberry]

Peter Majtan put together a good review of histograms and acquisition for Scarletuser too.

http://www.scarletuser.com/showthread.php?t=567

 

[Irri]

...If you underexpose Red by 1 stop you have effectively halved your dynamic range as the top 2048 values are sitting idle....

Is that technically correct? I can see that you are using half of the information, but half the dynamic range?

 

[Paul Didier Mogg]

Confusing Histogram and Image Clipping

Confusing Histogram and Image Clipping

Please excuse me if I'm hijacking this thread but my problem relates so closely to the subject that it seems appropriate.

Could anyone please explain to me why I'm seeing what looks like blatant clipping in the whites in the attached screen grab from RedAlert?

I thought I had exposed this image nearly perfectly, I "exposed to the right" without any clipping showing on the Histogram as you can see. Yet when I look at it in RedAlert I see clipping in the white dresses on my monitors. When output to ProRes (HQ) it's even worse. What am I not understanding? ...and if the information is in fact there how do I recover it? Doesn't the histogram show that in fact I could have even exposed a little more?

Any advice on this is gratefully received as I'm now totally confused.

Regards,

Paul Mogg

 

[Irri]

Could be that the RAW is clipped and you are bringing it down with the 160ASA. What happens when you look at it at 320ASA?

 

[Paul Didier Mogg]

Yes, at ISO 320 it looks like the attached picture.
But I shot it with the camera set at ISO 160, with no clipping showing on the Histogram or the zebras, that were set to 95%.
I thought that the Histogram showed the RAW sensor exposure data, and was to be regarded as God on the subject.
Have I misunderstood the whole thing?
What you seem to be implying is that unless you shoot as 320 ISO, the native setting for the sensor, you're wasting your time in using the Histogram as an exposure reference. If that is the case, does the False Color screen show true sensor exposure? If not that then what?
I've obviously got something very wrong here.

 

[Frank Mirbach]

Hmm, strange..

 

[Irri]

To me, it looks like the raw image has blown out highlights. Perhaps the monitoring at 160ASA still doesn't show you when you are clipping properly. I think the histogram in Red Alert is monitoring the final image so although it is clipped, the 160ASA setting pulls those values down from 100%

Try leaving it at 320ASA and using the exposure control to see if you can recover any more information in the highlights. Then try the DRX slider to try to recover detail.

 

[Frank Mirbach]

To me, it looks like the raw image has blown out highlights. Perhaps the monitoring at 160ASA still doesn't show you when you are clipping properly. I think the histogram in Red Alert is monitoring the final image so although it is clipped, the 160ASA setting pulls those values down from 100%

Try leaving it at 320ASA and using the exposure control to see if you can recover any more information in the highlights. Then try the DRX slider to try to recover detail.

It seems to be that way. Although it would make way more sense to make the histogram showing ONLY what is definitely captured in the RAW image that has been created untouched by the ASA setting, for it´s metadata only anyway.