Histograms in RawTherapee


(Paul Deverson) #1

I am a newbie to RawTherapee and would appreciate some feedback regarding the histogram. I have a Nikon D90 and have what I would have considered a well-exposed RAW image of a Gretag-Macbeth Colour Chart. If I apply a neutral profile, the histogram is spread evenly across the graph with roughly equal gaps at both ends. However, if I view it in the raw histogram, everything is shunted to the left hand side, with nothing on the right 50%. Does this mean that it is actually under-exposed, and that I should be exposing so that the right side is filled, and then adjusting the exposure compensation when processing?


#2

Hi @pauld
Could you upload the RAW image so that we could have a look at it?


(Paul Deverson) #3

AttachedDSC_0057.NEF (9.0 MB)


#4

Thank you, @pauld

I have taken a swift look at it, and I’d say that the photo is
under-exposed.

If I remember it correctly, a “properly” exposed/illuminated
Colour Chart such as yours should show values around L a b 96 0 0
in the white square, and about L a b 20 0 0 in the black square.

Have fun!
Claes in Lund, Sweden


(Glenn Butcher) #5

I opened you raw in RT, and I’m now looking at your raw histogram. If your image were exposed such that the top-right patch were blown, you’d see a spike at the right hand side in all three channels. That, and the RGB values from that patch in the Navigator are at 89-90% of the exposure range, so I’d say you might have one stop to play with. The horizontal scale of the histogram is logarithmic, so you need to keep that in mind when comparing data spreads at the right versus those on the left. I’d probably go just 1/2 a stop on this one, because when you shoot a color checker or other target, you really want to know what the camera woud measure at white, and blowing it through saturation really goofs up that information.

After messing with all this for a bit, I’ve stopped thinking of the concept of “under-exposure”. You can bracket-expose at the scene, bring those raws home and produce depictions from them that for all practical purposes look identical to casual observation. The data is going to get scaled at some point for regarding because the way its recorded by the camera doesn’t line up with the scale at which we view it. But, if you look more closely, you’ll see differences in the highlights that were clipped, and in the shadows where they were recorded at the range where your sensor struggles with noise.


(Paul Deverson) #6

Thanks, Claus. But won’t these values vary according to which profile you apply? I suppose what I’m driving at is: should I be using the raw histogram to determine exposure (i.e. push to the right as far as possible) or use a normal histogram after applying a profile (e.g. neutral)?


(Morgan Hardwood) #7

There is ~0.7EV room for improvement, which is nothing and not worth bothering.


(Glenn Butcher) #8

The thing you want to control with exposure is the information your camera actually records. Too little light, and your measurements of it are predominantly dark and noisy. Too much light, it overwhelms the sensor and its analog-digital converter, and all such are just clipped to an arbitrary value corresponding to saturation. Even the neutral profile scales the data apparently, after the exposure, so it confounds your assessment of where the light sat on the sensor.

Some scenes are just too much dynamic range for your camera, so now you have to make tradeoff decisions: clip those light sources or specular highlights, or push the rest of the image below the noise floor. Or, bracket three exposures and HDR them in post. Or, same three bracketed images, pick the one which trades off the thing about which you care less…

Oh, for the color checker, you do want to not blow that white patch if you’re going to use it to make a camera profile. The software that makes those needs to know what your camera thinks is “white”, not the arbitrary R=G=B that you get at clipped saturation.


(Paul Deverson) #9

Thanks to all for your help. My conclusion is that the ideal exposure is one that just prevents the highlights clipping using the raw histogram.