"Beware of the histogram display. It stinks."

[John Crellin]

Didn't proof read my text, apologies. By "bottom" I meant left.

Oh and I see I then went on to use confusing words like above - by which I meant to the right - above in the sense of brighter or to the right. In a sense there isn't much important information in a histogram in the up/down direction as far as I can see - apart from whether there are any pixels at a particular luminance at all.

Should have left it to the glossary reference :-(

But the point I was trying to make in the final bit is that I do find the D70 will often tend to leave space at the bright or right end of the plot if the dynamic range is low - which I am sure is what leads to all the worry about dull results. All it needs is a tweak in the curves control but it happens.

If it cuts off at the bottom you have lost some detail down in the
shadows and will get black for parts of the picture - no D70 frame
to show you where this is (unless I missed something !)

This should read "cuts off on the left hand side". The histogram
is a plot of the number of pixels (vertical scale) for each
luminance value (horizontal scale).

As a secondary function, if there is space above the plot with
nothing in it then the D70 has underexposed and, if you don't want
to post-process, then try again with more exposure. But in my
experience this only happens when the range of the scene is low and
you haven't actually lost anything but you will get the famous D70
dull picture without post - processing !

Changing the exposure will move the whole graph left or right, not
up or down. If the histogram ends a quarter of the way from the
right hand edge but runs up to the left hand edge with lots of
pixels there, then you have probably underexposed.

Bob Meyers

 

[Leon Wittwer]

As usual, Ken blows things out of proportion.

However, one thing always seems to get left out in
histogram/luminance discussions. Everyone assumes that the
luminance is calculated as the sum of the photosites. But is this
assumption true?
Perhaps it is a simple sum of the color channels,
but it's not out of the question to consider that Nikon may have
programmed some color channel compensation into the equation. In
fact, to do this correctly, it would have to also consider WB since
this affects the overall temperature of the image. I would not be
surprised if this were the case.

While I don't know have specific knowledge about the D70, it is common practice to calculate the luminance pixel for pixel from a jpg of the image generated according to the current settings of the camera including the WB setting. The image may or not be a thumbnail which is a decimated version of the jpg image. If a thumbnail is being used, then small bright spots may not show well as they may be partially averaged out. It depends on how things are done.

I would love to hear from someone who has technical knowledge on this.

--
Leon
http://pws.prserv.net/lees_pics/landscapes.htm

 

[Leon Wittwer]

IMO the purpose of the histogram is more oriented towards helping
you pick up a better exposure so as to maximize the use of
available CCD data bandwidth. The d70 has also a "highlights" mode
which make overexposed areas in the picture blink (no matter which
channel is overexposed): this is truly the one most useful to look
at overexposure.

Does the blinking occur if a color channel is blown out but the luminance is not? That is often why it is easy in some cameras to blow out a color channel while the luminance is not blown out.

If a picture has a high luminance contrast, you may decide to
accept a few overexposed pixels. The histogram does not tell you
which areas of the pic will lose details. The highlights mode will,
and you can make an informed decsion on the importance of these
areas to your picture and accept the best compromise.

Thierry

--
Leon
http://pws.prserv.net/lees_pics/landscapes.htm

 

[MMuddler]

The histogram
is a plot of the number of pixels (vertical scale) for each
luminance value (horizontal scale).

Changing the exposure will move the whole graph left or right, not
up or down. If the histogram ends a quarter of the way from the
right hand edge but runs up to the left hand edge with lots of
pixels there, then you have probably underexposed.

you've faded to black!

Bob Meyers

Bob,

I think you are right in correcting the previous poster. The point you make is that an exposure change shifts the distribution of pixels to the left or right. You can get them bunched up in the middle if "properly exposed" and the subject has little contrast or you can spread them out with higher contrast lighting or subject matter. Or you can pile them up at the right edge of the histogram by over-exposing...etc.etc. but the area under the histogram ought to remain the same.

So how come the amplitude and shape of the histogram can vary with exposure?

Question... is what you see in the histogram affected by the curve in the camera? Should be... but I don't know. The following then is conjecture.

How far the pixel count ( that's the vertical scale) shifts when exposure is adjusted is probably a function of the slope of the curve. ... steep slope makes drastic change in output luminance for increment of observed luminance.... If you had a truly linear curve (input output relationship) the histogram would simply shift left or right... except for the pile-up at the extremes. If you have an inflection point in the curve then you'll see bunching up or thinning out of the histogram. Check what happens when you vary exposure on one of the serious S shaped curves.

The point of the histogram is to tell you how much of the image falls in the various brightness ranges and most importantly how many pixels have hit the wall and are beyond post processing recovery.

So what if you are shooting in rapidly changing outdoors situations and can't see the results in the LCD and can barely make out the histogram and don't have time to pause to look?

I would like the camera manufacturers to consider a viewfinder warning (oh, but there's too much in there already that I don't interpret!) that tells me when I have x pixels that have hit my white end of the scale ( or X% of my total pixels) in pure white....(sort of like blinking highlites on LCD) but I'd like to be able to set x or X% so I can be tolerant of a certain level.. Could do the same on the dark end with y pixels in black or Y% of total in black. (No blinking shadows) Now if you could do that by color, we'd be able to know which channel was in danger of blowing highlites/burying detail.

Marabou Muddler

 

[danandjan]

It'll be too time consuming for me to look at 3 different histograms or it'lll look too crowded to view all 3 in one screen. In a user interface point of view, I'd rather use just the one histogram IMO.

Reading a review of the D70 by Ken Rockwell, I came through these
writings.

"Beware of the histogram display. It stinks. The histogram display
is defective in design and therefore less than useless because its
inaccuracy can lead you to make heinously overexposed images that
still read AOK. The problem simply is that the histogram display of
the D70 (just like the D1H and D1X) only displays the green
channel."

How far is this true? Is this a real problem when you are shooting
in the field? As far as I know, histograms are very important to
check the accuracy of exposures. Then, why did Nikon do this to the
D70 (if what Mr Rockwell says is true!).

I'm no expert, so may be someone may throw more light on this
issue! I've just ordered a D70 and this histogram thing is worrying
me a bit!

--
Nikon D70 with kit lens, 24mm, and 50 mm lenses.

 

[TOF guy]

Leon

Indeed the blinking will occur if only one channel is nearly overexposed or overexposed.

Since luminance is a weighted average, you can have one channel overexposed but the average will not exceed its maximum value, and no pixel will be counted at the far right of the histogram.

Thierry

 

[muskiemouth]

If provided it would have to be configurable. Some folks want one, some want three, let them eat cake.

Reading a review of the D70 by Ken Rockwell, I came through these
writings.

"Beware of the histogram display. It stinks. The histogram display
is defective in design and therefore less than useless because its
inaccuracy can lead you to make heinously overexposed images that
still read AOK. The problem simply is that the histogram display of
the D70 (just like the D1H and D1X) only displays the green
channel."

How far is this true? Is this a real problem when you are shooting
in the field? As far as I know, histograms are very important to
check the accuracy of exposures. Then, why did Nikon do this to the
D70 (if what Mr Rockwell says is true!).

I'm no expert, so may be someone may throw more light on this
issue! I've just ordered a D70 and this histogram thing is worrying
me a bit!

--
Nikon D70 with kit lens, 24mm, and 50 mm lenses.

 

[joe field]

I quite often use the blinking highlights feature because if there is just a small portion of the shot with questionable highlights, it is hard or impossible to see it on the right side of the histogram.

It would be nice if Nikon let us see the histogram with or without the blinking highlights.

Concerning evaluating a shot taken. I find more frustration with the display in terms of being able to see the image in sunlight and with the poor zoom feature for checking critical focus.

Joe

 

[DaveHutch]

Well, my methodology may be severely flawed (and I'm sure I'll be shot down if it is), but I thought I'd try a little experiment.

I made three images on my PC that each showed a graduated full screen going from black to red, black to green and black to blue.

I photographed these at correct exposure and also at +5 then looked at the histograms in PaintShop Pro and on the camera LCD

In Paint Shop Pro, the channels were individually very blown out indeed at +5and if you viewed the single colour channel histogram this was obvious.

If however you viewed the greyscale histogram, the exposure was right on the money.

now for the LCD, this definitley seems to show the greyscale histogram, not the green channel. It's histogram matches the greyscale histogram in PaintShop Pro in every case.

Lastly, I cannot get the Highlights display to flash even if a single channel is SEVERELY blown out.
Go on, try it...
--
Dave H
First thing I do in the morning is smile.....get it over with

 

[MMuddler]

Well, my methodology may be severely flawed (and I'm sure I'll be
shot down if it is), but I thought I'd try a little experiment.
I made three images on my PC that each showed a graduated full
screen going from black to red, black to green and black to blue.

Dave, Your experiment has some interesting results! Can you show us these three images with a file so we can print them to run the test ourselves? Not sure i understand your graded color chart.

I photographed these at correct exposure and also at +5 then looked
at the histograms in PaintShop Pro and on the camera LCD

Correct exposure metering from what part of the target chart?

In Paint Shop Pro, the channels were individually very blown out
indeed at +5and if you viewed the single colour channel histogram
this was obvious.
If however you viewed the greyscale histogram, the exposure was
right on the money.
now for the LCD, this definitley seems to show the greyscale
histogram, not the green channel. It's histogram matches the
greyscale histogram in PaintShop Pro in every case.
Lastly, I cannot get the Highlights display to flash even if a
single channel is SEVERELY blown out.

That's the surprising observation... even at +5 !!??

Go on, try it...
--
Dave H
First thing I do in the morning is smile.....get it over with

--
Marabou Muddler

 

[DaveHutch]

Well, my methodology may be severely flawed (and I'm sure I'll be
shot down if it is), but I thought I'd try a little experiment.
I made three images on my PC that each showed a graduated full
screen going from black to red, black to green and black to blue.

Dave, Your experiment has some interesting results! Can you show
us these three images with a file so we can print them to run the
test ourselves? Not sure i understand your graded color chart.

the image goes from fully red on the left and graduates to fully black on the right.
I can post some on my website but not until this evening when I get home

I photographed these at correct exposure and also at +5 then looked
at the histograms in PaintShop Pro and on the camera LCD

Correct exposure metering from what part of the target chart?

I used the centre of the image where I assumed it was roughly going to take the whole thing into account, but I used matrix metering

In Paint Shop Pro, the channels were individually very blown out
indeed at +5and if you viewed the single colour channel histogram
this was obvious.
If however you viewed the greyscale histogram, the exposure was
right on the money.
now for the LCD, this definitley seems to show the greyscale
histogram, not the green channel. It's histogram matches the
greyscale histogram in PaintShop Pro in every case.
Lastly, I cannot get the Highlights display to flash even if a
single channel is SEVERELY blown out.

That's the surprising observation... even at +5 !!??

I know, the PSP histogram reads fully blown in whatever colour but the LCD never flashes! Odd eh?

As a simple test, take a fully red folder, book, whatever and place it close to the camera, expose properly for the background and blow the red out with a flash. That may work too

--
Marabou Muddler

--
Dave H
First thing I do in the morning is smile.....get it over with

 

[DaveHutch]

having said all that, I may not get a chance to upload the files tonight. if not, I'll start a new thread on Monday with the links
Meanwhile, have a try yourself:

Shoot a red PC screen at 0.0 and +5 exposure, do the same with blue and green then look at the histograms in PS and on the camera.
You should see what I mean.

it would be nice to have someone else do this anyway to see if they get the same results
--
Dave H

 

[MMuddler]

After looking at dave's experiment I started something (intended to test Dave's no-flashing-highlites result) that gave me some observations along the way that I don't quite understand yet......regarding the LCD histogram versus the PSCS histogram (viewed thru Levels at RGB view).

I see a couple of things described below but i'd be interested in hearing if you can see them for yourselves with my set up since I'm not sure if I've messed up...

I created full screen color patches in photoshop... pure red, green and blue. e.g. red green blue 255,0,0 etc. And then I shot them on my calibrated monitor with the D70 fullscreen out of focus at ev +0 and ev +3, +4 and ev +5. I've got Foto's provia loaded.

first ... the camera's LCD histogram places the metered exposure for a pure blue screen differently (more to the left) than pure red or pure green. That is observed when comparing o ev shots for the three pure colors as well as comparing +3 shots of those colors. It would seem blue is underexposed by at least 1 stop.... red and green only a partial f stop. One the one hand I'm not surprised that metering isn't the same, but on the other hand I am somewhat surprised. [BTW, the highlites flash on all these monochrome targets ... but have to look again to see if the ev threshold is different.]

second unexplained observation....as expected, the RGB histogram in PS Levels shows a small spike on the right edge and large spike on the left edge for each of these monochrome targets presumeably saying one channel is blown and two channels are black( these are the histograms of the created target screens...not the camera capture of them...maybe I'll will look at that later). Contrast the multispiked PS RGB histogram with the one-eyed LCD histogram.

I later created Red Green, RedBlue and Blue Green mixes...... 255,255,0 and 255,0,255 and 0,255,255 targets ....confirmed when looking at individual channel histograms. And I expected to see the RGB histograms for each of these blended targetss to show a large spike at the right edge for the two colors that were saturated and a smaller spike on the left edge for the one channel at 0. But that observation was only seen for RedBlue and RedGreen. The BlueGreen combination produced RGB histogram spikes of ???? equal magnitude on the left and right edges. The one black channel was Red and it appears to have as much weight as the maxed out Blue and Green channels combined. Hmmm...sure that is easily explained but that caught me by surprise.

Before i look at the RAW output I want to get past the two issues above. Any thoughts or explanations?
--
Time to get dressed and go to work!

Marabou Muddler

 

[Gillis Danielsen]

Atleast if the d70 is like on my 10d it's not very precise, the only thing you really see is the blinking highlights. but otherwise it's very hard to judge if you hit the spot with the exposure because the lcd varies alot on the surrounding

But speaking about ken rockwell... i haven't been taking him seriously since that raw vs. jpg post and a bunch of other stupid things he writes... of course he can have his oppinion and maybe he is right sometimes too, but the way he writes them is to make everyone else look like imbessilles and idiots and he being the guru who knows it all...

 

[Iliah Borg]

I've done that test using red, green, and blue zonal color separation filters.

L = 0.299 * R + 0.587 * G + 0.114 * B

same as composite histogram in PS

--
no text

 

[MMuddler]

I've done that test using red, green, and blue zonal color
separation filters.

L = 0.299 * R + 0.587 * G + 0.114 * B

same as composite histogram in PS

--
no text

Explains shift in the camera LCD histogram for red, green and blue exposures but doesn't address the PS RGB histogram observations I made in Levels on the blended colors in my muddled experiment.... why Red is so strong vs blue/green.

Different issue than camera's algorithm for luminance.
--
Marabou Muddler

 

[Iliah Borg]

Explains shift in the camera LCD histogram for red, green and
blue exposures but doesn't address the PS RGB histogram
observations I made in Levels on the blended colors in my muddled
experiment.... why Red is so strong vs blue/green.

CRT red is not pure red

--
no text

 

[MMuddler]

Explains shift in the camera LCD histogram for red, green and
blue exposures but doesn't address the PS RGB histogram
observations I made in Levels on the blended colors in my muddled
experiment.... why Red is so strong vs blue/green.

CRT red is not pure red

--
no text

Didn't rely on CRT red in the issue I raised about PS histogram.

The 2nd observation I made in my muddler's experiment post was solely based on what was created in PS and did not rely on photographing a monitor representation of Red or the Blue Green blend. The question arising from the PS Levels RGB histogram was based on an internally created BlueGreen screen (w/o and Red)... viz, 0,255,255. .... why the PS histogram of that blend showing a spike on the left edge for Red that seems to have as many pixels in it as the BlueGreen spike on the right.
--
Marabou Muddler

 

[Iliah Borg]

The 2nd observation I made in my muddler's experiment post was
solely based on what was created in PS and did not rely on
photographing a monitor representation of Red or the Blue Green
blend. The question arising from the PS Levels RGB histogram was
based on an internally created BlueGreen screen (w/o and Red)...
viz, 0,255,255. .... why the PS histogram of that blend showing a
spike on the left edge for Red that seems to have as many pixels in
it as the BlueGreen spike on the right.

Now, I see Blue-Green is Cyan; Cyan is opposite to Red - so what you see is a bug

--
no text

 

[Iliah Borg]

For better illustrating the point in the previous post - open an empty CMYK file, filled w/white - and look at histograms of Cyan, Magenta, Yellow, and Black channels.
--
no text