Camera Histograms Lie

Yes, the histograms your camera shows you are "a lie."

 

At least they are if you have a raw workflow. A raw file is just a massive array of ones and zeros. It has no "meaning," and it cannot be displayed until it is interpreted and defined in relation to something else. This requires defining a "color space" in which to display the data among other things.

In order for your camera to display anything it created a jpeg "thumbnail" file. This file is compressed and has the sRGB color space and that is part of the problem.

To demonstrate some of the possible issues I used a test image from Cambridge in Color which was designed to test histograms.

Here is what the original file looks like when opened in PS.

 

Note the histograms. Luminance and red are not clipping while green and blue are. Also note the heavy bias to the left of all of the curves.

I took this image and enlarged it in order to fill the viewfinder of my camera (D800) and I took a picture of it. I set the jpeg camera settings so that the camera display most closely matched the original image.

Here is the raw file opened in PS.

 

The histograms have changed with blue now being the only channel clipping. This is obviously due to my not having captured the entiire original image. It may be interesting to note that this is not what the histograms looked like when initially opened. There was an "alert" that the histograms being displayed were "cached data" and needed updated which means "the histograms" are actually defined in the file based upon the camera preview. This is what they looked like immediately after updating the histograms to actual (non-cached). It is also what they looked like when opening the file in LightRoom. (Lightroom appears to ignore the cached data).

And here is what the histograms looked like on the camera.

 

 

These two "previews" are for the exact same image! The banding in this image is due to photographing a screen with a slow refresh rate. In person it looked much better. Note the hard color transitions (they were not as bad in person). This is probably due in part to the camera having a 256 color display. But it is also possibly due to the displayed image being a small/compressed file with the sRGB color space optimized for the camera's display (i.e. only having 256 colors). 

And note the histograms. They look nothing like they do for the raw file. Blue is the closest, but now all three channels are showing hard clipping. And all of the channels "weight" have shifted right. Heavily in the luminance, red and green channels. And also note that the blue channel is not clippiing as much. Why?

Here it seems that some "perceptual bias" has been applied. Since blue is 11% of an exposure shift left, red is 30% so it gets a little more emphasis, and green is 59% so it gets pushed right hard.

It also seems that the recommended exposure index (REI/ISO) is being applied. Most cameras have a lower actual ISO than is reported by the camera and this is done to help avoid the harsh highlight clipping characteristics of digital sensors. Basically, the ISO is "set lower" because the manufacturers recommend slight undereposure. So now all of the channels are showing clipping (overexposure) and the luminance channel is getting closer.

And notice the luminance histogram, it also looks nothing like it "should." It looks more like a stackup of the RGB channels. Had I pushed this histogram right in order to "expose to the right" based upon the luminance histogram or the highlight warnings (blinkies) I would have severely clipped the blue channel. With highlight warning turned on I had no blinkies for luminance, but I did have some for all of the channels which matches the camera histograms. But it does not match the actual raw file. This is probably where "the extra headroom" in a raw file comes from.

Apparently, at least some Canon's have even worse behaviour as they will show a pixel as being clipped if any of the colors reads above ~250 and not 255. Here is a link to similar testing of a Canon 5D (item 9) and also mentions the 1Ds. 

Because this is really "a way of thinking" in how the manufacturers want to interpret/display the information I would suspect all cameras to have some extent of the same issues.

What do the manufacturers have to say about this? Here's the disclaimer in the Nikon owner's manual: "camera histograms are intended as a guide only and may differ from those displayed in imaging applications." I.e. differ from the real ones. Canon doesn't even bother to mention it.

So what does this mean really? I don't know, don't trust the histograms? I've gotten by for a very long time without using the histograms and simply judging the image itself as displayed. And by knowing I've got an extra couple of stops in the shadows and an extra stop in the highlights that are in the raw file (approximately). In fact, I don't review most of my images after capture at all. I suppose that is why I find it so "alarming" to learn that using the histograms could be so misleading.

And while anything climbing the right edge of a histogram is "clipped" and lost data, generally clipped highlights (luminance) is worse than having a single color clipped. And even some highlight clipping may be acceptable. For instance, I expect a hard specular highlight on chrome to be clipped/blown.

 

Or, you can use a Uni-WB(white balance) or Magic Lantern if you have a Canon EOS.

When using a Uni-WB your image review will look green, but the histograms will be much more accurate. I have found that using the "camera neutral" profile with everythig but contrast set to zero(middle), and contrast set to the maximum to be the most accurate.

 

But let's put this back into perspective. This test was designed to expose "issues" and the test image is not characteristic of the levels one would find in a typical scene. At least not all of them together. 


UPDATE: The particular behaviour seems to vary between brand and even camera model. My D4 has much less of an issue although it isn't 100% accurate either.

FaceBook  Twitter