1D X and D4: At the Limits of Exposure

In this second synthetic test of our 1D X and D4 evaluation, we are going to look at dynamic range, or more simply put, we’ll test how much usable detail we can push up from the darkest areas and pull back from the brightest areas in the RAW files of these cameras. Previous articles in the series:

If you are a photographer, you can probably guess which half of the image below offers a greater chance of recovering a usable image:

Left half: 5 stops underexposed, right half: 3 stops overexposed

Left half 5 stops underexposed, right half 3 stops overexposed

We had a pretty good idea, too, but decided to test it anyway and in the process, compared how good the cameras in our test are at recovering information from under- and overexposed areas of the RAW image.

But first, why RAW files and not JPG-s? In a JPG file, the brightness/intensity of each primary color can have a value between 0 and 255, that is 256 different values. In a RAW file, however, these cameras can store up to 16384 different values for each primary color. This means that when we make a JPG file in camera or on our computer, we throw away quite a bit of information that the camera was actually able to see.

For example, let’s take a simplified look at highlights. In a very bright part of the picture, the camera might have recorded intensity values in the range between 16320 and 16384 – that’s 64 different levels of brightness that could constitute some image detail. When creating a JPG, the camera will have to squeeze the 16 384 different values it can tell apart into the JPG range of 0-255. This means that all the values between 16320 and 16384 (our 64 different levels of brightness) will be assigned the same value of 255 – and we’ll have an area of uniform color and no detail whatsoever.

If we shot JPG only, there is no way to tell all these “255″ values apart from each other and therefore no way to restore the initial 64 different shades that the camera saw in that image area. This is what happens when we overexpose a bright area of the image by two stops and try to get the lost information back from a JPG – an area of solid color where the properly exposed image showed detail.

If we have the RAW file of the overexposed shot available, we can tell the computer to “squeeze” the 16384 unique values into the 0-255 range in a different way, so that the differences in the brightest areas still remain visible to a degree. From the overexposed RAW on the right, we still managed to get some of our nice fluffy detail back.

It is important to realize that the RAW file advantage is not just about saving improperly exposed images – there are many everyday scenes that have such contrast / dynamic range (ie very different brightness values) that even RAW files are unable to store all the different values and maintain detail in the lightest and darkest parts of the image.

So what we are looking at here is how much usable information each of the cameras is able to record in the shadows and highlights that the photographer can then bring into the final print or a JPG on the web.

Pulling back 2 stops of overexposure

It seems that two stops of overexposure are generally recoverable – all the cameras are very close here and it’s hard to find any meaningful differences. There is one exception – ISO 50. We knew that ISO 50 is supposed to have very little highlight headroom and now we got to test it. The difference between ISO 100 and ISO 50 is indeed substantial. The 1D X suffers the most here, losing all detail in the lightest areas; D600 does best, followed by 5D Mark III. But even the better cameras lose detail and even color at ISO 50.

In fact, the results are very similar to pulling back 3 stops of overexposure at ISO 100 and this makes sense – ISO 50 is supposed to be an extended ISO setting and basically an ISO 50 shot is overexposed by a stop at ISO 100 and then pulled back by the camera.

You can also observe that overexposing and then pulling the exposure back reduces apparent noise – the overexposed and adjusted ISO 12800 shot looks much cleaner then the correctly exposed one. However, this is no magic cure for noise – if you have the shutter speed and aperture to overexpose by two stops at ISO 12800, you should be shooting at ISO 3200 anyway – and these shots look quite similar. But by exposing normally at ISO 3200, you are much less likely to blow out highlights.

Tom: This post will be the long one so just brief comments from me. Pulling back 2 stops is the limit today. We decided not to move up by thirds of the stop – its enough to know your limitations. 5D Mark III performs well. Not the first place thow.

Pulling back 3 stops of overexposure

If two stops seemed fairly easy to pull back, overexposing by three stops has ruined some parts of the test image beyond recovery. No more fluffy details for the 1 D X at +3 stops. The D4 is a tiny bit better, but not by much. Interestingly enough, things start getting better for the full-frame Canons and the D4 when we increase the ISO. An overexposed shot at ISO 800 has more highlight detail than a similar shot at ISO 100. We don’t have an explanation for that, you are welcome to help us out in the comments.

You’ll also notice that at this extreme overexposure, we are losing color, with 7D being the worst offender in this regard. This is because if we overexpose a color so badly that the brightness of all its primary color components exceeds the maximum value the sensor can record (clipped in all channels is a nicer way of putting it), there is no information to tell us what the real color mix and thus the actual color was. So when we try to recover this part of the image, all we get is neutral grey. The D600 seems to get the most color out of this situation, at least when looking at the yellow patches that have no color on other cameras, except maybe the D800.

Overall, D600 seems to lead the pack at ISO 100, with the 5D3 not too far behind. D4 and D800 make up the next tier but the 1D X and 7D are really struggling here. At higher ISOs, the Canons gain ground, particularly the 5D3. At ISO 1600 it seems to retain more highlight detail than the D600.

Tom: D600 leading. To my personal surprise 5D Mark III holds its valuable second place. 1D X burns in our purgatory test with bright flames like 7D.

Pushing up 5 stops of underexposure

This is where it gets interesting and we will see some big differences between the cameras. Why try to push such an extreme underexposure? Because we can, or at least the best cameras here can. The enemy we face here is not clipping, however, but shadow noise.

When the 5D3 and the D800 were launched in early 2012 and initial comparisons were being made, it was a real eye-opener to see the kind of noise advantage that the Nikon had in low-ISO shadows. But would the D4 (which has a somewhat different sensor from the D800) hold a similar advantage over Canon’s top-of-the-line 1D X? See for yourself. To sum it up, the D4 is not quite as amazing as the D800, but still pretty close; the 1D X is better than the 5D3, but not nearly good enough to take the fight to the Nikons.

As for the other cameras we tested, the D600 follows right in the footsteps of the D800 and its low ISO shadow noise performance is clearly superior to any Canon on the market today. The 7D is the oldest model here, but somewhat surprisingly, it seems less noisy in this scenario than the 5D3. It is, however, plagued by very obvious pattern noise (the vertical stripes) that is resistant to normal noise reduction techniques and looks very bad in actual pictures.

The advances in sensor technology by Sony and Nikon that Canon is unable to match at the moment, start to cancel out at higher ISOs. I’d say the break-even point is at around ISO 3200 for the D4 and the 1D X, may-be a fraction of a stop earlier for D600 vs 5D3.

I would like to emphasize that this is not just about rescuing underexposed images. It is also about being able to use the full dynamic range of the camera to protect highlights by exposing for them and knowing that there will still be usable information in the shadows that can be pushed up.

A word about the extreme ISOs as well – there really is less noise at ISO 50, so it makes sense to use it sometimes, but only when you know you won’t have highlights to worry about. As for the highest ISOs, well, there really wasn’t much point in pushing five-figure ISOs by five stops, but hey, if you ever wondered what ISO 6,5 million would look like with current technology – here’s your answer.

Tom: At ISO value 6,5 million I would take D4. But seriously, at ISO 50-400 it is Nikon (also landscape photographers) domain. I liked we took those JPG references here to show that all your hard work with RAW files can pay you off.

In conclusion

Having spent some time on the Internet during the past year or so, it was no surprise to see the Nikons dominating the shadow pushing competition. It is probably the most obvious image quality differentiator between the current full-frame cameras, but its relevance is, of course, up to the individual photographer. I don’t think its essential, but it sure can help in some photographic situations and during more aggressive post-processing.

It was also known beforehand that overexposure is more likely to deliver a fatal blow to an image than underexposure, but it was good to quantify that a bit. In our experience, two stops over can mostly be recovered, but an extra stop on top of that can become an insurmountable problem. The ability to recover from underexposure is more dependent on camera and ISO, but at least at lower ISOs, there is more room for error.

As for specific cameras, we confirmed that they are close when dealing with overexposure, but the Nikon shadow noise advantage remains evident even in the flagship models. The D600 was once again a strong contender, both in shadow and highlight handling, especially considering its status as an entry-level model. On the Canon side, the 1D X was clearly better at dealing with underexposure, but the 5D3 did better with highlights, so no clear winner there.

Tom: I think this is the most interesting part of our ISO test. I also like to thank Martin, who cared to write a more detailed comments about this theme. So far so good – I have a dream and passion, Martin has brains to sort these messy things out. I hope you liked it too.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>