Questions About Histogram on 5D Mk3

Started by Pedr, November 21, 2015, 09:22:12 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Pedr

According to this thread: http://www.magiclantern.fm/forum/index.php?topic=12096.0, when I select EV units for spot metering, the value is:

Quotethe EV from saturation (overexposure)

And I assume this is true of the use/display of EVs elsewhere in the overlays.

At 100 ISO if I set the histogram to show dynamic range (Overlay > RAW EV Indicator > Dynamic Range), it gives me the value of D11, which tallies up with what I've read about the 5D Mk3 which is that it has 11 stops of dynamic range.

However, If I set the histogram to show an ETTR Hint  (Overlay > RAW EV Indicator > Show ETTR Hint) and cover my lens, the maximum value that is displayed is 9.3EV, suggesting that the difference between no exposure (because the lens is covered) and overexposure / over-saturation is only 9.3EV.

Additionally, with the lens covered, the histogram displays data over the first three EV bars. With no light entering the camera, I would expect a spike at (or around) the left edge of the first EV bar.

So why does the ETTR hint max out at 9.3EV when the camera has a dynamic range of 11EV, and why does the histogram show 3EVs of data when no light is entering the camera?

a1ex

The noise in the image tricks ETTR here. The hint is computed from the brightest pixels on the histogram, which are quite a bit above one standard deviation (which gives the noise floor).

Let's check the math:

- At DR = 11 EV, you have about 7 units of noise (this is one standard deviation). White level is roughly about 15000, so DR = log2(15000/7) = 11.07 EV.

- If ETTR would meter on the 99.7th percentile (a rough approximation), that level would be at 3 standard deviations above black. So, the hint would be log2(15000/7/3) = 9.48 EV.

Pretty close, no?

BTW, in the absence of light, the correct value for the ETTR hint would be... infinity. And the histogram will show - guess what - the noise.

Recommended reading: http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

Pedr

Thanks for the detailed explanation and that link looks great.