Interpreting Histogram Information

arthurgoh wrote in post #2285635
That's interesting. I guess that implies that the vertical axis (amount of pixels) is less than the total pixel count of the sensor, and therefore once a certain number of pixels with the same tonal range exceeds the axis the histogram peaks 'off the chart'.

I wonder if it's possible, or useful even, for cameras' onboard computers to dynamically scale the histogram so that all its pixels show on the chart even if the entire frame is filled with just one tone. Perhaps we can even experiment with different camera models with same sensor size to see if the max pixels on the vertical axis is constant.

That would be good. Almost like with digital audio. To scale or 'normalize' the waveform.

1st spike on the right is the sky

2nd spike is the wall

play with your levels, move the white slider around and see what turns white.

HOSTED PHOTO
please log in to view hosted photos in full size.

HOSTED PHOTO
please log in to view hosted photos in full size.

arthurgoh wrote in post #2285635
That's interesting. I guess that implies that the vertical axis (amount of pixels) is less than the total pixel count of the sensor, and therefore once a certain number of pixels with the same tonal range exceeds the axis the histogram peaks 'off the chart'.

I wonder if it's possible, or useful even, for cameras' onboard computers to dynamically scale the histogram so that all its pixels show on the chart even if the entire frame is filled with just one tone. Perhaps we can even experiment with different camera models with same sensor size to see if the max pixels on the vertical axis is constant.

You wouldn't want it to scale - because that way you'd lose all the nuances whenever there's a major spike.

OK cool, so the bottom line is that there is just too much information to be displayed but there's nothing wrong with seeing that topping out type of spiking?

Glenn NK wrote in post #2285427
I thought that a flat top on a histogram spike simply indicated that a lot of pixels captured information at that particular dynamic level, and that the information is there, but simply can't be charted on the limited vertical range of the histogram.

Yes, now that I've seen the histogram, that's exactly what's happening here. The designer of the software made an explicit decision to chop off the top of tall spikes.

In this case, what the flat top indicates is a low-contrast picture (or more accurately, a picture with all of the information squeezed into one or more small tonal ranges).

arthurgoh wrote in post #2285635
I wonder if it's possible, or useful even, for cameras' onboard computers to dynamically scale the histogram so that all its pixels show on the chart even if the entire frame is filled with just one tone. Perhaps we can even experiment with different camera models with same sensor size to see if the max pixels on the vertical axis is constant.

There's no question that it's possible; it would be a trivial change to the firmware. What would be really handy, though, would be if you could change the vertical scale with the multicontroller. That way you could zoom in for detail or zoom out to see the true shape of that clipped-off spike. (This, too, would require trivial firmware changes. Somebody should suggest it to Canon!)

TMR Design wrote in post #2285946
OK cool, so the bottom line is that there is just too much information to be displayed but there's nothing wrong with seeing that topping out type of spiking?

Iv never had a problem with it, and Iv got plenty of histograms that are topped out like that.
To the best of my knowledge its distribution across the X axis that you need to be aware of.

My understanding of the histogram is each Pixel gets its own spot on the graph, which is scaled 0-255 across the X axis.
The Y axis is then scaled in such away that you can get a good idea of how the pixels are spaced across the X axis, while still having some idea, of how many are grouped at each point.
The ideal histogram would have a Y axis as tall as the number of megapixels in your camera. In even a 1MP camera it would show as a very small, tall thin graph, BUT, you would be unable to properly see the X axis and the distribution of data on it unless you had a similary very tall thin LCD on your camera.
So the Y axis is compressed, and the top is cut off.
The data is still collected by the camera, but any extremely high concentrations of the same luminosity are cut off when displayed on the histogram graph in oder to keep the graph a viewable size.

So what your seeing in the example photo is a limited tonal range, comprised of two lighter colours (in this case greys) occuring in two very high concentrations, with a clearly visable but small contrast differnce.
There are no whites in high enough concentration to be visable in the graph, and there is a small number of darker colours (in this case dark greys) spread thought out the image.

As long as it's not bunched up to either the left or right, you're cool. EC, or playing with the curve in PP if you shoot RAW might let you draw more info out of that part of the curve, sometimes without losing the ends.

As long as you're not shooting polar bears in snow storms....

Titus213 wrote in post #2286097
As long as you're not shooting polar bears in snow storms....

Or black cats in coal cellars


Have you seen this:

http://www.luminous-landscape.com …standing-histograms.shtml

http://www.cambridgein​colour.com/tutorials.h​tm

These tutorials on the histogram really show it.
Regards
Don

arthurgoh wrote in post #2285635
That's interesting. I guess that implies that the vertical axis (amount of pixels) is less than the total pixel count of the sensor, and therefore once a certain number of pixels with the same tonal range exceeds the axis the histogram peaks 'off the chart'.

I wonder if it's possible, or useful even, for cameras' onboard computers to dynamically scale the histogram so that all its pixels show on the chart even if the entire frame is filled with just one tone. Perhaps we can even experiment with different camera models with same sensor size to see if the max pixels on the vertical axis is constant.

My understanding (doesn't mean I'm right) is that the camera or editing software counts the number of pixels for each level of brightness (dynamic range), and plots the brightness on the x-axis and the pixel count vertically. If the count exceeds the vertical axis range, the plot is chopped off.

It would be interesting to see the total count, particularly at the extreme dark/left and light/right ends of the plot. "I know I blew the highlites, I want to know how badly".