Optimum digital exposure: the birders case

[Rory]

Further to the discussion on the optimum digital exposure, I would like some help on a specific case.  Photographing birds with white feathers often presents the situation where the feathers can be extremely bright when at specific angles to the sun and images with blown feathers are suboptimal.  The following example was shot with a 7DII, 600mm + 1.4X in manual mode at f/5.6, 1/1600 sec and ISO 800.  Light conditions were changing continuously and the subjects were hopping about with considerable energy.  I was adjusting the shutter speed and ISO on the fly, basically winging it (bird metaphors fit I guess :*)).

Here is the full frame of the image from DPP 4:



Zooming in to show the feather blowout.  As you can see, a small portion of the feathers do not have any detail, although I have pulled the highlight slider down in DPP.



I used DPP, with as shot settings (all noise reduction turned off) to convert the image to a TIF and rendered in Lightroom, where I also adjusted the highlights.  As you can see, there is still some blown highlights.



Since Adobe has yet to release ACR for the 7DII it could well be that the DPP rendered TIF is insufficient to allow LR to render detail in the highlight area.  To try to determine if this was the case, and to check if I had over exposed, I took a look at the image in Raw Digger, which also suggests that there is some blowout according to the OvrExp indicator.  I am a little confused here, as the max for the G and G2 channels is 13,256.  Is this the max for the 7DII?  I thought it used 14 bits which would be around 16,000.



The red channel looks better.



Here is the Raw Digger histogram for the entire image:



Here I have selected the portion of the image where most of the over-exposure is occurring:



And here is the histogram for the selected portion, where I can see the blowout in the green channels.



Hopefully I still have your attention at this point. 

• Is this image over-exposed or do you think it is possible to render some detail for the brightest feathers?
• If the image is over-exposed, how would you recommend metering to avoid feather blowout?

Thanks
Rory

[Rory]

I think I can answer one of my questions, about why clipping occurs at 13,256.  I think Canon changes the clipping depending on the ISO, and in this case the ISO was 800.

[Fine_Art]

Bracketing has saved my bacon on many a bird shot.

[telyt]

Whenever possible I'll take a few test exposures and check the histogram for any overexposed channels, adjusting exposure (manually) as required.  Doesn't work too well for tiny stuff jumping in & out of light unless you decide which light is the one you want and wait for the bird to jump into your chosen light.

[Rory]

Bracketing has saved my bacon on many a bird shot.

That's a great idea.  I've been thinking about it.  Turning it on and off seems a little clumsy in the Canon bodies.  Is there a trick?

[Rory]

Whenever possible I'll take a few test exposures and check the histogram for any overexposed channels, adjusting exposure (manually) as required.  Doesn't work too well for tiny stuff jumping in & out of light unless you decide which light is the one you want and wait for the bird to jump into your chosen light.

That is what I normally do but I usually end up with significant under-exposures.  This was not such a big deal when I'm shooting Nikon with the D4 or D800E, but with the 7DII I'm finding that you have to try to eke out all the ETTR possible.  I'll have to take a look at the exposures on a number of my shots to see if there is some type of sunny f/16 type rule I can use.  It gets tricky with changing ISO, since the exposure ceiling drops as the ISO goes up.

[telyt]

Perhaps the 7DII isn't optimum for this task.

[kirkt]

Place a white feather in various lighting conditions that occur when you shoot and acquire a series of exposures at various ISO - see which combination of ETTR and raw converter gives you the latitude you require to balance your exposure parameters and still maintain/reclaim your highlight detail.

I am finding more and more nowadays that, for specific exposure issues like yours, exposure is as much determined by the destination for raw conversion as it is by the camera itself.

In the end, you may just have to expose for the extreme highlights if they are that important and let the rest of the tonal range fall where it will, remapping it in post.  One advantage of having white feathers blow out is that, if there are still detail data left in the red and blue channels, you can reconstruct the highlight data by placing the good channel into the blown channels at the same tonal level, keeping the white neutral (R=G=B) and reintroducing detail.

kirk

[Rory]

Thanks Kirk.  I agree - testing is in order.  Good idea about the feather.

[Ligament]

Rory, great post!

[ErikKaffehr]

Hi Rory,

Use "Logarithmic" scale on the Y-axis and things will be more clear.

This info may also be useful: https://www.youtube.com/watch?feature=player_embedded&v=NdLjU70xhdQ

I may add that I don't shoot a lot of birds, but I also feel they are a difficult case. I may also add that I don't foresee giant steps on the sensor side. They are decently close to physical limits. Making them larger helps a bit, but than you need a longer lens. Twice the sensor diameter, half the noise and eight times heavier lens. To that comes the $s.

Best regards
Erik

Ps. Well, there are rumors about non Bayer sensors, from Sony and Canon, and they may be more efficient.

[Rory]

Thanks Erik.  Very informative.

[Fine_Art]

Here is an example of a specular effect which looks odd. With higher DR, this would have been ok. If you look at the rear leg, it has some water droplets which blew out in the sunlight. Since the bird was walking forward following prey, bracketting would not get the legs atthesame angle.

The top is cropped, otherwise as shot.

[Guillermo Luijk]

ETTR makes sense when there is something to gain in terms of noise. This usually happens in medium and high contrast scenes. With a modern sensor and under these situations (low DR scene compared to sensor's DR), trying to achieve ETTR is a risky business and moreover useless.

Always leave some highlight headroom by applying negative exposure compensation as a regular basis, no matter if your JPEG displays a bit dark or you need to push exposure in the RAW developer a bit. Your sensor can deal with that shadows noise, your highlights will be preserved, and in addition you'll be able to shoot faster!.

ETTL era?

This is how a not so new APS-C camera (Pentax K5) behaves when 6 stops of underexposure are applied in a low DR scene:

(left camera JPEG, right RAW+6EV in pp)


RAW histogram of the previous capture:



Of course this is an extreme example and some parts in the jacket had a far from desired SNR. I'm talking about 1-1.5 stops of underexposure vs perfect ETTR.

[deejjjaaaa]

This is how a not so new APS-C camera (Pentax K5) behaves when 6 stops of underexposure are applied in a low DR scene:

indeed... sized for a social media.

[Dave Ellis]

I took a look at the image in Raw Digger, which also suggests that there is some blowout according to the OvrExp indicator.  I am a little confused here, as the max for the G and G2 channels is 13,256.  Is this the max for the 7DII?  I thought it used 14 bits which would be around 16,000.

Rory I found similar results with a Canon 600D some time ago. It seems to me that the clipping level of the ISO amplifier following the sensor is somewhat lower than the maximum input level that the A/D converter can handle. Hence the clipping point is somewhat lower than 16,384 which would correspond to the A/D converter's maximum input. Not sure why there would need to be such a large margin but I dare say there is a reason.

You certainly have some clipping there in the raw file but I wonder whether ACR will be able to do something with it. I've read that under certain circumstances, actual clipping like this can be recovered in ACR by the use of some smart algorithms. I draw a distinction here between actual clipping and clipping in the camera developed jpeg which seems to set a white point that gives a safety margin, and this can be readily recovered in ACR because the raw signal isn't actually clipped.

Dave

[Rory]

You certainly have some clipping there in the raw file but I wonder whether ACR will be able to do something with it. I've read that under certain circumstances, actual clipping like this can be recovered in ACR by the use of some smart algorithms. I draw a distinction here between actual clipping and clipping in the camera developed jpeg which seems to set a white point that gives a safety margin, and this can be readily recovered in ACR because the raw signal isn't actually clipped.

Dave

I was able to improve it slightly with ACR but ended up doing a little cloning as well to make it look natural.

[Guillermo Luijk]

Not reaching the end of the ADC bit scale in the RAW data is common. The Canon 5D used to clip at 3692 rather than 4095. That's a fairly good amount of levels wasted* for a 12-bit camera (and yet we have to add the non-zero bias offset typical from Canon).




* But probably only cameras that do so can enjoy the whole DR of the photosites, while others saturate the ADC input signal losing (a bit of) highlights information instead. I see clipping occurring at this initial stage as a reduction of the effective full well capacity of the sensor.

[Fine_Art]

Not reaching the end of the ADC bit scale in the RAW data is common. The Canon 5D used to clip at 3692 rather than 4095. That's a fairly good amount of levels wasted* for a 12-bit camera (and yet we have to add the non-zero bias offset typical from Canon).




* But probably only cameras that do so can enjoy the whole DR of the photosites, while others saturate the ADC input signal losing (a bit of) highlights information instead. I see clipping occurring at this initial stage as a reduction of the effective full well capacity of the sensor.

Is there any real reason for this? Such as preventing blooming in nearby pixels? Surely a large matrix meter could set the exposure to 99% at the brightest metered spot of the frame?

[Dave Ellis]

Just yesterday I had a look in RawDigger at some files from my Nikon D610. For ISO values from 100 through to 800, the clipping point (in the green channel) was the same at 15,780. At ISO1600 it rose to 16,273. These values are quite close to the maximum for a 14 bit A/D converter (16384).

This suggests to me that the components in the sensor (photosites, amps, A/D converters) are all well matched.

The fact that the clipping point doesn't change from ISO100 to ISO800 suggests to me that the full well capacity of the photosites is a bit higher than the amp clipping point.

Dave