Author Topic: Exploring options for HDR metering through the viewfinder  (Read 2496 times)

Audionut

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 3657
  • Blunt and to the point
Exploring options for HDR metering through the viewfinder
« on: May 01, 2014, 07:29:21 AM »
Related discussion:  https://bitbucket.org/hudson/magic-lantern/commits/70f59636d89d1491aa66e17860f99bf6ebb7b4dc



Brief evaluative metering refresher.

http://media.digitalcameraworld.com/files/2012/06/Metering_mode_photography_cheat_sheet.jpg

Canon metering attempts to consider every photon in the scene, combine all of the values obtained, find the average, and meter that average for a midtone (13% in the case of Canon).  From this midtone point, there are defined points where saturation (over exposure) occurs, and the point where noise becomes greater then the signal.  Dynamic range.

Evaluative metering places some priority at the focus point.  So if we place the focus point over an area of the scene that is predominantly darker then the average, the Camera will attempt to meter this area closer to a midtone. And the same occurs when we place the focus point over an area of the scene that is predominantly brighter then the average.  This is useful if we decide we want to place priority on the shadows or the highlights.  In other words, let's place (some exposure) priority in these areas, so they don't get crushed to black, or over exposed.



Consider this scene.



The red dots indicate the focus points used to evaluate this scene (5D3).  Note:  The indicated focus points may not be 100% accurate, use your best judgement.
With aperture and ISO locked, the camera recommended these shutter settings for exposure.

Quote
1/400
1/250
1/640
1/800
1/500
1/200
1/320

We can see a spread pattern of 2 EV for this scene.  Remember, the indicated exposure settings are for a midtone. 
In this case, the darkest exposure recommendation 1/800 (middle top focus point), still wasn't enough to capture the highlights without any saturation.  It was close, only slightly saturating the green channel in the brightest parts.  Note:  Most recent Canon cameras will saturate around 3.5 EV from the midtone.

Also, in this case, the brightest exposure recommendation was 1/200 (far right bottom focus point).  Here, the exposure feedback has exposed the shadow point, as a midtone.  So here, the point where SNR=1 is some 7 EV below this midtone exposure.

So, we have 1/800 with a saturation point around 3.5 EV above this, and 1/200 with a bottom limit around 7 EV below this point.  If we add all of this together, we have 3.5 EV (above the darkest midtone metering ), 7 EV (below the brightest midtone metering), and the 2 EV between the metered spread pattern, for a total of 12.5 EV of required dynamic range.

Recommended exposure settings.  1/800 with a 3 EV recovery ISO.  In this case, we still had some slight over exposure in the highlights.  However, on scenes where a focus point only covers the highlights, I would expect that we should see sufficient headroom.

Now, I don't pretend for a second that this will provide the user with 100% accurate exposure.  However, if I wanted 100% accurate exposure, I would use existing functions (AETTR, bracketing). 

Unfortunately, these existing solutions are somewhat slow.  Currently, the only way to sufficiently meter for large dynamic range scenes, fast, is to meter for the highlights, using the newly developed Auto dual ISO.  Or previously, metering for the highlights with some pre-set dual ISO setting. 

Both of these metering practices, require the user to be able to meter for the highlights.  By implementing exposure decisions based on automatic calculations of evaluative metering, from various focus points, we make another successful attempt at turning our expensive DSLR's into Point and Shoots   :P

All of this, in the vain hope, that ML can control the focus point.

Another option, would be to spot meter the highlights and the shadows.  However, I expect this to be rather difficult and complicated to code, and the speed slows somewhat, where you probably should just use AETTR.