Friday, September 21, 2012

Experiments on Nikon D600 Exposure

UPDATE:  I posted about this on DPReview, got some helpful responses suggesting that there are issues with my shooting method.  That DPReview thread is here if you want to check it out.  I have done follow up experiments that are inconclusive as to the advice given in that thread.  The D600 still appears to overexpose.  I will do a followup post about it.

UPDATE 2: 2nd set of experiments here

I've shot with a number of cameras and they all tend to have different exposure tendencies. For example the Pentax K100D and Fuji S5 attempt to avoid blowing highlights, often producing straight-out-of-the-camera (SOOC) images that appear underexposed.  The Nikon D80 doesn't seem to be very concerned with highlights and produces SOOC images where the subject is correctly exposed but the image often has blown highlights such as in the background. Then there are cameras such as the Nikon D70 and Nikon D3 which in my opinion have a balanced exposure (taking into account the limits of the sensor's dynamic range).

In this post, I took some test shots with the D600 to evaluate whether the D600 has a tendency to overexpose or underexpose. All shots were taken in raw and processed in View NX2.

I took a shot of some backlit foliage below an overpass. It was sunny and bright outside. In aperture priority mode at 100 ISO, the camera chose an exposure of f/4, 1/100. Given how sunny it was I would have expected a much higher shutter speed, perhaps 1/800 or 1/1600 (it was quite sunny). Anyhow this is the resulting image, and its histogram.

In ViewNX2, I applied a highlight recovery of 100 to see how much I could avoid from blowing out. This screenshot shows the clipped highlights AFTER the highlight recovery.

As you can see the sky is totally blown. Sometimes that really can't be avoided especially with older digital sensors. But if you look at the histogram you'll see that in this case none of the shadows were clipped at all, so there was room to accommodate at least some of the highlights. This seems weird - almost as if the camera was exposing to protect the shadows rather than the highlights. That is the right thing to do when using negative film, but with digital it's quite the opposite - you ought to protect the highlights because digital sensors have tremendous latitude in the shadows but limited range in the highlights. When highlights are clipped they disappear suddenly, as though a hole were cut out of the picture (there's no gradient, unlike with negative film).

I took a shot of a tree and some more foliage, under similarly sunny conditions. Again I used aperture priority. I was at 100 ISO, f/4.  The camera chose a shutter speed of 1/160. This was the result and
its histogram:

Look carefully and you can see the very tall spike at the highlight end. Uh oh. And here are the clipped highlights after 100 highlight recovery.

Well, at least the D600 is consistent. It looks like the camera again decided to just forget about the sky. As with the previous test shot, no shadows were clipped.

I took a shot of this black Ferrari. It was in open shade, specifically in a covered parking lot but facing the sunlight just a few feet away. I used aperture priority ISO 100, f/4. The camera chose a shutter speed of 1/3 secs. (WTH?) Here is the result.

Check out that histogram. On the highlight side, a bunch of pixels are crowding apartment number 255,255,255. Meanwhile on the shadow side, it's pretty lonely. Let's check out the clipped highlights.

Whoooaaa! Really? That's just crazy and pretty hard to justify.  Applying 100 highlight recovery, let's see how much we can get back.

That's still a pretty big chunk to give up, in my opinion. I tried turning Active D-Lighting on. Active D-Lighting is supposed to preserve highlights by underexposing slightly, then brings up the underexposed midtones and shadows by tweaking the tone curve.  (Canon's version is called Highlight Tone Priority). The adjustment to the tone curve doesn't matter to raw shooters like me because we prefer to apply our own adjustments, but ADL is useful for influencing the automatic exposure to avoid blowing highlights.

I turned ADL on Auto. The shutter speed was still astonishingly slow so I bumped ISO to 1600 (note: this reduces dynamic range). Even at 1600 ISO the shutter speed was just 1/60. That seemed crazy to me --
it's not like it was night time... :-? Anyway, I took a shot. Here's the result.

And here are the clipped highlights.

Much improved but still, lots of clipping on the red and green channels (yellow is the intersection of red and green). With highlight recovery there were even fewer clipped portions:

The thing is, I don't like to use too much highlight recovery because at some point the image starts to look unnatural. I wanted to see if I could coax the D600 into preserving all highlights. I turned ADL to
Extra High. Here is the result.

And here are the clipped highlights with ADL at Extra High:

Still showing many clipped highlights.  Yes they are mostly just specular highlights but still, I expected that there would be none with ADL at Extra High.  Now with an older sensor I might have said, perhaps the dynamic range of the scene was just wider than what the sensor can handle.  But if you look at the shadow side of the histogram, it looks like there's still room to back off the exposure.  Anyway, here's the same shot, with highlight protection at 100:

Finally, in an effort to really stamp out any blown highlights, I tried just dialing down exposure compensation by -1.3 EV.  Here's the shot.

And here are the clipped highlights:

Even with no highlight recovery there are barely any clipped highlights.  Yes the image appears underexposed judging by the grayness of the white van, but I know I am going to adjust that in post-processing.  This is just the starting point for me, and I don't want to have any cutout holes in my canvas.

My hypothesis from these test shots is that the D600 chooses an exposure that attempts to get a correct exposure of the subject, without necessarily looking at the rest of the frame. In thinking about this, I guess it might be ok if you have no plans whatsoever to postprocess the shot, and all you want is a SOOC image that looks as good as possible without any editing.  I mean if you were to judge the Ferrari shots above, the -1.3EV shot looks underexposed, and the default shot arguably looks better if you don't look closely at the van's lost highlights.

For someone who processes his images as part of his normal workflow, this algorithm is not ideal. For me, the SOOC image is almost never the endpoint. It's just the beginning and I like an image that has as much available relevant information as possible.  I'm not going to try to capture EVERYthing.  I mean if the sun was in the shot I'm not planning to dial down the exposure til I see sunspots. :)  But generally, I would like to keep the details for everything that is relevant to the shot then I can adjust them to my preference during postprocessing.

What makes this tragic in the case of the D600 is that from earlier experiments, it appears to have tremendous shadow recovery abilities.  So, even if a shot is underexposed, it appears the D600 should be capable of correcting that underexposure without too much drama.  The D600's exposure behavior doesn't take advantage of that capability.

One factor to consider is that ViewNX2 is not that great compared to Lightroom 4. It's possible that there's actually recoverable detail in those areas in Lightroom 4 even though you can't recover them in ViewNX2, in which case the D600's exposure might be ok.

For now, though, the D600's exposure seems too aggressive for me.  I will tone it down by using the extra high or high ADL setting and I will be watching that histogram closely.

And if any of you readers have suggestions or corrections to my technique please chime in through the comments.  Or email me at info AT   Meanwhile I'll be working on part 2 of the review.

Nikon D600 Hands-On Review Part 1
Choosing the Exposure for Dynamic Range