are histograms logarithmically displayed? (relevant for expose to the right)

[spacediver]

Been reading about expose to the right, and the only way this makes sense to me is if the histograms being used to guide exposure are displayed logarithmically.

Is this the case?

 

[jsecordphoto]

What

 

[spacediver]

What I mean is whether the X axis on the histogram is displayed in linear or log units.

 

[fjrabon]

yes, but code assignment is linear, thus the theory behind ETTR

 

[spacediver]

yep, the sensors are linear. It was just very confusing reading these ETTR articles because I was assuming that the histograms were linear. The articles skipped that essential piece of information. Thanks for the clarification.

 

[Alexr25]

The X axis of the histogram displays the pixel value (0 - 255 in each channel for an 8 bit file), the Y axis displays the number of pixels of that value.
The pixel value will have had a gamma correction applied to it (see http://www.poynton.com/notes/colour_and_gamma/GammaFAQ.html ) so yes the X axis is effectively logarithmic,.

 

[spacediver]

ah ok makes sense. I process all my RAW images in Matlab, and when I look at the data I use a regular, linear light histogram. That's what threw me for a loop, I've little to no experience working with the histogram on the camera itself. And Poynton is the man

 

[TCampbell]

Histograms are linear. They are very simple. The X axis represents pixels from black (RGB value of 0, 0, 0) to white (RGB value of 255, 255, 255 if you're thinking JPEG... in RAW the cameras have a higher number of bits per channel... usually 14 bits.)

The Y axis simply represents the number of pixels of that tonal value. You can display the histogram as a simply luminosity value (the most common) or you can display it showing separate RGB channels.

 

[Ysarex]

The histograms displayed on camera as well as the histograms displayed by post processing software do not directly reflect the sensor exposure data. In the case of the camera they are derived from the RGB photo produced by the camera processing software. If you're concerned about ETTR which involves exposing the sensor then your not looking at graphs of the sensor exposure data -- a complication of some consequence.

As Tim noted most camera histograms are luminosity histograms of the camera's image processor output. As such they will not indicate single channel clipping.

Joe

 

[Designer]

Exposing to the Right - Digital Photography School

Digital Exposure Techniques

ETTR Exposed Digital Photography Review

 

[fjrabon]

Histograms are linear. They are very simple. The X axis represents pixels from black (RGB value of 0, 0, 0) to white (RGB value of 255, 255, 255 if you're thinking JPEG... in RAW the cameras have a higher number of bits per channel... usually 14 bits.)

The Y axis simply represents the number of pixels of that tonal value. You can display the histogram as a simply luminosity value (the most common) or you can display it showing separate RGB channels.

They're linear if you define them based on RGB values, but isn't that just because RGB values are logarithmic (hence why RGB 127, 127, 127 is 18% grey, not 50%)?

 

[spacediver]

yep, RGB is generally gamma corrected, and depending on the nature of the exact opto-electronic conversion function used by the camera, the relationship between the pixel values and scene luminance is more or less logarithmic.

Seems to me the best method is to evaluate exposure in each channel in linear light (i.e. work with RAW).

 

[unpopular]

^^ The sensor's electrical property is, for the most part, linear. This is why gamma correction is necessary to start with.

---

Typically the value axis is linear, and the pixel count axis is either linear or logarithmic. I don't think I have ever seen a histogram with a non-linear value axis - but, it's been a while since I've used software that had the option (and when I did, I kept it in linear scale) so I could be wrong.

As for ETTR, just place the brightest region of the scene at the edge of the histogram without over-exposure and place shadows in raw processing using curves or levels. This is regardless if the histogram is linear or logarithmic.

ETTR is often over-complicated through imprecise and non-photographic hedging. Figure out how many stops above middle grey will clip, and just spot meter the brightest region and increase exposure by just under that amount.

The only time you don't want to do this is under very low contrast scene which would require a severe adjustment to compenesate.

 

[Ysarex]

^^ The sensor's electrical property is, for the most part, linear. This is why gamma correction is necessary to start with.

---

Typically the value axis is linear, and the pixel count axis is either linear or logarithmic. I don't think I have ever seen a histogram with a non-linear value axis - but, it's been a while since I've used software that had the option (and when I did, I kept it in linear scale) so I could be wrong.

As for ETTR, just place the brightest region of the scene at the edge of the histogram without over-exposure and place shadows in raw processing using curves or levels. This is regardless if the histogram is linear or logarithmic.

ETTR is often over-complicated through imprecise and non-photographic hedging. Figure out how many stops above middle grey will clip, and just spot meter the brightest region and increase exposure by just under that amount.

The only time you don't want to do this is under very low contrast scene which would require a severe adjustment to compenesate.

Good old GIMP.

Joe

 

[Ysarex]

yep, RGB is generally gamma corrected, and depending on the nature of the exact opto-electronic conversion function used by the camera, the relationship between the pixel values and scene luminance is more or less logarithmic.

Seems to me the best method is to evaluate exposure in each channel in linear light (i.e. work with RAW).

Indispensable: RawDigger Raw Image Analyzer RawDigger

Joe