Spot Metering & Histograms

[Peter Ryan]

Hi Peter,

The game changer is when photographing scenes with back lighting - then the dynamic range required can jump by 3 stops or more in the blink of an eye - and it's here that exposure becomes more critical as 4 stops of reflectivity - plus 3 stops of backlighting - plus a couple of stops of shadow detail take you to 9 stops and most cameras will only get to around 11 or 12 - ...........

so you're getting pretty close to the noise floor and risk shadow noise when you raise the shadow levels in PP, if the shot is under-exposed to start with.

Thanks Colin,

I am interested in the last sentence, which I have broken out of the paragraph to highlight. You have mentioned before in a post the phrase "close to the noise floor" and I was not up to understanding the answer at that time and I am not sure I am now but I think I am closer to trying to understand this point so I ask - can you explain the last sentnce to me?

[inkista]

One other trivial point to keep in mind is that the histogram is not generated off the sensor or RAW data image (if you're shooting RAW), it's being read from the already-compressed JPEG preview of the image. So your histogram is not 100% accurate, and you may in fact have more latitude on either end than you think you do. And the in-camera settings for contrast can also affect the dynamic range of the JPEG preview and therefore the histogram. Reducing the contrast can sometimes "gain" you more latitude with processed data, which is why HDSLR videographers are using those types of presets to "increase" the dynamic range they have to work with in the final footage.

[Peter Ryan]

I understand what the histogram reusults from but your comments on reducing contrast are interesting. Thanks for contributing.

[Colin Southern]

Thanks Colin,

I am interested in the last sentence, which I have broken out of the paragraph to highlight. You have mentioned before in a post the phrase "close to the noise floor" and I was not up to understanding the answer at that time and I am not sure I am now but I think I am closer to trying to understand this point so I ask - can you explain the last sentnce to me?

Hi Peter,

No worries ...

It's about signal to noise ratio ... if a sensor captures a tone pretty close to the maximum it can handle then it will be a lot of signal, and very little noise. But as things are captured that aren't as bright, the noise stays the same, but the signal level drops and the signal to noise ratio decreases. Carry on and there gets to be a point where signal and noise are roughly the same, and it's impossible to tell what's signal and what's noise. Or put another way, the further the captured tone is from the max, the more noise will be mixed in.

For most cameras, the difference between the two - at base ISO - is about 12 stops - so if you're only photographing a reflective scene (4 stop dynamic range) and you under expose be a stop (pretty typical of what matrix / evaluative metering will do) then the darkest tone recorded is only 5 stops down from the max, and still 7 stops clear of the noise floor - so still pretty clean. If you want to tap into a couple more stops of shadow detail then that's still not a problem (still 5 to spare) ... but when you start getting backlighting (ie active light sources, not just reflected light) then that can easily take up the top 3 stops of dynamic range, and everything else gets shoved down - closer to the noise floor.

Thing to keep in mind is that although the camera is capturing 12 stops, monitors only display 6 and prints only 4 (ball park figures) - so to bring some of the captured detail into a range where we can see it, we have to compress it onto the top 4 to 6 stops (ie "boost the base") which is what the fill light slider does in Photoshop (and to a lesser degree the brightness slider). So when these "low tones" are brought into view - if they're too close to the noise floor - the noise is dragged up too. In essence the black clipping point slider then becomes the demarcation between what is considered signal and what is considered noise - normally, if the shot isn't underexposed too much to start with, and the scene doesn't have too much backlighting / active light sources etc then this isn't an issue - but it becomes more of an issue with scenes that have a relatively high dynamic range and/or you're shooting at a high ISO (because you lose 1 stop of dynamic range for each stop of ISO you increase on the camera).

In practice what this means is two things ...

1. When shooting into the light, push the exposure to the absolute limit if there is shadow detail that you need to reveal, and

2. When shooting at high ISOs, be aware of the reduced dynamic range capability of the camera, and expose accordingly.

Does this answer the question?

[Peter Ryan]

Hi Peter,

For most cameras, the difference between the two - at base ISO - is about 12 stops - so if you're only photographing a reflective scene (4 stop dynamic range) and you under expose be a stop (pretty typical of what matrix / evaluative metering will do) then the darkest tone recorded is only 5 stops down from the max, and still 7 stops clear of the noise floor - so still pretty clean. If you want to tap into a couple more stops of shadow detail then that's still not a problem (still 5 to spare) ... but when you start getting backlighting (ie active light sources, not just reflected light) then that can easily take up the top 3 stops of dynamic range, and everything else gets shoved down - closer to the noise floor.

Does this answer the question?

Yes I think I have it, thank you Colin. As I said I am at that stage where I can start to understand some of these more technical issues (for me). So for clarity, when you talk about the maximum signal a sensor can handle you are talking about the white clipping point and everything falls away from there down some 12 stops to the noise floor and the closer the two get the more noise increases relative to the signal because noise is constant and the signal weakens.

[Wirefox]

One other trivial point to keep in mind is that the histogram is not generated off the sensor or RAW data image (if you're shooting RAW), it's being read from the already-compressed JPEG preview of the image. So your histogram is not 100% accurate, and you may in fact have more latitude on either end than you think you do

Thats interesting...thanks Kathy. I am at a loss as to why that should be though since I would have expected the raw data to be easier to translate rather than going to the trouble of re-translating the jpeg data. Speed and cost i assume.

Thanks for joining in. No, I was saying the histogram measures the brightness of the whole scene but the spot meter measures only part of the scene so how do those who actively use spot metering use the histogram to make exposure fine tune adjustments with the EV compensation when you are only measuring off a small part of the scene and you don't know if that part is actually mid tone. Yes using a grey card or white card and adjusting accordingly works but I hear many say meaure off some object withing the scene itself and I pretty sure most cannot pick a mid tone to measure off. So if they picked the wrong tone then looking at the histogram will not hel them understand they got it wrong.

Ah I get what you are saying now but I still can't see why spot metering would cause you to read the histogram differently. This is where I get lost because if I were using a grey card I would spot meter off of that. However if the scene was complex in its light distribution I would spot meter off the bit I want to be best exposed and make manual adjustments to compensate for any over or under cooking in other areas of the scene. Typically though something has to give in a scene like that and with spot metering I decide which area of the scene gives and which gains. As far as I can tell this control is not possible with matrix metering.

Someone is going to tell me I have been doing this all wrong...but the results seen reasonable

[inkista]

Thats interesting...thanks Kathy. I am at a loss as to why that should be though since I would have expected the raw data to be easier to translate rather than going to the trouble of re-translating the jpeg data. Speed and cost i assume.

Speed would be my guess. To offer a live histogram feature, it probably makes sense to merely grab the data fed to the LCD, rather than the full sensor feed. Chewing through and graphing the full 21MP of a 5D Mk II sensor is going to be a lot harder to do fast enough for live feedback than the <1 MP of data that's displayed on the LCD.

And for the same reason, compressing the data before displaying it on the LCD probably speeds things up. Hardware compression is fast. It's software compression that's slow.

[Wirefox]

Great in depth technical knowledge you have Kathy. That all makes perfect sense

[arith]

Personally I am a matrix metering man and the histogram tells me how the tones in the scene are recorded in camera.

Those of you who shoot in Spot metering mode (or Centre Weighted for that matter) cannot look at the histogram in the same way otherwise why not just meter in matrix metering in the first place.

My question then is, how do you read the histogram if you are metering in Spot mode, what does it tell you, not about the scene but how you have metered the spot - can you tell if it was sucessfult or not?

I use simple logic; the aperture and time just allow a certain amount of light in, irrespective and independent of how you meter. So if I want nothing to be blown I meter on the brightest spot to be 2 stops overexposed.

That normally will result in something being blown because a) you couldn't see the brightest spot and b) spot metering isn't really spot but averaged over about 7% of the centre.

So after some trial and error I use 5/3 stops overexposed.

The resulting image displayed on the back of the camera is a jpeg with a strong curve even if you shoot raw, which I always do. So it will show warnings.

However 2 stops can be recovered in ACR, so not too worry eh. The longer you can expose the more information you record up to blowing the pic altogether.

The only way I deviate from this is if the EV range is greater than 8EV shown on the meter which is probably greater than 11 EV in reality, then I just feel sick and think what about that zone thing or can I find a way to do HDR.

[arith]

Maybe, I think I got your point.

The base line of the histogram (the area of the graph to the right) must hit about 1/2 of the last quarter of the graph (if the histogram is divided into 4 parts) OR must hit 1/3 of the last slice of the graph (if the histogram is divided into 5 parts). Something like this:



This graph was taken from the wonderful book of David Sizer "Captured by the Light". Figure 4 is the adjusted exposure based on the histogram from Figure 5.

I don't find either of these histograms to be showing good exposure; whilst there isn't clipping of lowlights, more memory is allocated to record highlights and therefore more information could be recorded by doubling the exposure in the top and even quadrupling it in the bottom.

I just read about memory allocation somewhere, it looks like your making a meal out of the really easy.

[Dave Humphries]

I was saying the histogram measures the brightness of the whole scene but the spot meter measures only part of the scene so how do those who actively use spot metering use the histogram to make exposure fine tune adjustments with the EV compensation when you are only measuring off a small part of the scene and you don't know if that part is actually mid tone.

Hi Peter,

The LCD histogram shows the overall exposure curve for the entire scene captured, it'll either be correct, too far left (hence probably under exposed), or too far right, maybe even clipped (i.e. over exposed).

I think we all accept it does that regardless of;
the meter mode used (spot, centre weighted or average/matrix). or
the reflectance of what you metered off (grey card, white card, something in the image)

You look at the histogram and just decide whether you need more or less exposure, end of story.
You then adjust the exposure accordingly; if manually exposing, by means of shutter speed (usually), or iso, or aperture, or by using + or - EC if semi-auto exposure.

It is an adjustment based on the overall result alone.
Not whatever tone, grey card, etc. you metered off

An analogy may help: you put a CD in the CD player, you set the volume at #4, since that's what you normally have it at, but if the ambient noise is high, or the track faint (some are), you won't be able to hear it well, so you turn it up in response to what you heard, not by looking at the volume knob, or turning it a set number of degrees because that's what you normally do, or someone told you too. Just by ear.

~ but I hear many say meaure off some object within the scene itself and I'm pretty sure most cannot pick a mid tone to measure off. So if they picked the wrong tone then looking at the histogram will not help them understand they got it wrong.

What did they get wrong Peter? How do you expect them to see it?
All they can see is an overall histogram, with a lump, hopefully in the middle.

Hi dave,

Thanks for contributing. I understand the histogram tells you the same information but as in the above post, the histogram does not tell you if you chose an incorrect tone to measure off whereas with matrix metering you can make EV adjustments based on the blinkies and the histogram to adjust your exposure.

No it doesn't.

As above, if we have all accepted that the histogram doesn't change with metering mode - why do you feel the histogram any less valid at judging exposure overall and setting +/-EC when in spot mode than when in matrix mode?

Regarding the tone level metered off or shape/area of the scene they metered off and any weighting applied, how/why does that matter?
Let's suppose the camera had a little cross-hair aiming cursor and a digital read out of the RGB or Luma value where the cursor was positioned, how would that help? Like knowing your on volume 4?
Surely it is better to know the overall image isn't over or under exposed?

The metering mode used and the tone you might meter off in spot mode are just the use of tools to help get a good exposure first time, just as setting the volume to #4 or turning the knob 55 degrees is. Then you take the picture (or hit Play on CD) and judge the results; by histogram/blinkies or ear, NOW the initial guesstimate is irrelevant, as is how you got it (degrees or numbers), what matters is whether you turn the knob (shutter speed or volume control) clockwise or counterclockwise, and how far.

Can't help feeling we're at cross purposes here

[Colin Southern]

I don't find either of these histograms to be showing good exposure; whilst there isn't clipping of lowlights, more memory is allocated to record highlights and therefore more information could be recorded by doubling the exposure in the top and even quadrupling it in the bottom.

I just read about memory allocation somewhere, it looks like your making a meal out of the really easy.

Steve,

Don't confuse "good exposure" with "no gaps" at the right hand end of a histogram". Yes - if the histogram touches the right hand side then the maximum amount of information is captured - BUT - it also means that the tone is recorded as a HIGHLIGHT ... and if the tone really ISN'T a highlight then this can and often does create problems trying to restore the correct tones in post processing.

Case in point ... If this was a sunset scene then without a doubt it's wasted a lot of potentially useful information - but if these were closeup shots of a models face then they would be correctly exposed and the histogram is as expected.

You can only ever correctly evaluate a histogram in the context of the image you're capturing.

[Peter Ryan]

Thanks everyone for your input. I understand what you are all saying and it has clarified in my mind the query I had but I feel I have not explained myself all that well trying to get my point across. Never mind the answer is well covered from all the disucssion.