My Canon EOS 5D

Dynamic Range

In this post, I'm going to look at the dynamic range of this camera. Rather than look at it in controlled conditions, which I'm not equipped to do anyway, I'm going to compare it to my two other cameras, a Nikon coolpix 2500 and a Nikon coolpix 4800. This won't give hard numbers, but it will provide a comparison between those cameras at a practical level.

Background

A bit of background before I get technical. My first camera was the 2MP 2500. I took that little camera everywhere with me. Over the years, I took thousands of photographs with it and never seemed to get a badly exposed one. Even in extreme cases, like on super-clear, summer days at high noon. Sure the shadows were horribly unflattering, but there were no parts of the image blown out and also none that were too dark to show detail.

Scoot forward a few years and I decided to upgrade to a 4MP camera. I picked the 4800, learned how to use it (I thought) and then went out to take pictures. Another summer, another super-clear, sunny, summer day at high noon and I decided to take some pictures. After all, the outdoor band was there playing, I was there listening, and I had my camera with me. Snap, snap, snap. When I got back and looked at the pictures I was astounded at how bad they were. The sunlit areas were all blown out and over exposed while at the same time the shaded areas were so dark they didn't have much detail left. I was really disappointed, the 2500 wouldn't have had any problems at all with that lighting, and I almost got rid of the camera assuming the problems it to be it's fault.. but luckily I learned better.. sort of.

What Went Wrong

After some talking to friends and reading on the Internet, I finally figured out what went wrong. I concluded that the problem was partly with the camera, but also partly with me – it was a dynamic range problem. When I went out shooting that day, I had set the ISO all the way to the top of it's range and had switched the color space from standard to “vivid”. I don't think I could have come up with settings for capturing the smallest dynamic range possible if I had tried. Not that the 4800 has particularly good dynamic range in the first place I've since learned, but those settings made it even worse. But why was it so bad? Before we tackle this question, let's take a short side trip and understand dynamic range.

What Is Dynamic Range

Put simply, dynamic range is the range of light values the camera can capture. In other words, how light and how dark can parts of the picture be when the camera can still capture detail in both. Typically, the lower end of the range of most cameras is pretty similar, so any dynamic range limitations are seen most easily at the high end of the range. Limited dynamic ranges are not necessarily a bad thing, photographers have been knowledgeably choosing chemical films with smaller dynamic range for years – slide films, and especially the extra-saturated slide films, have less dynamic range than negative film. Typically, the photographer decides where to apply their available dynamic range and sets their exposure accordingly.

For digital cameras, it turns out that dynamic range has everything to do with the size (square area) of the pixels on the CCD. Think of a CCD as an array of tiny light buckets. When the bucket is empty, you get black, when it is all the way full, you get white. Bigger buckets take more light to fill. The dynamic range of a CCD relates to how much light it can get before those buckets fill. Different settings on the camera can also change the cutoff where black and white occur on the buckets, making a slightly full bucket read as black, or a not-quite-full bucket read as white. Different cameras have different base dynamic ranges because the different CCDs used in them have different sized buckets. When one packs more pixels into the same size CCD, the buckets are necessarily smaller, and therefore fill faster when given the same amount of light.
Relating back to those two cameras, the 4800 has smaller buckets, which fill faster and with less light, so it couldn't capture the very bright spots. The metering electronics in the camera was putting the available dynamic range in the middle of the brightness range of the scene (where skin tones typically are) and therefore letting the dark regions get too dark and the light ones overexpose. Not a bad strategy actually, since most digital point-and-shoot cameras are used for snapshots of people. In my case, my selection of ISO400 was raising the amplification on the readings of the buckets, so even if they didn't get all the way to full, they would still read as white, meaning some of the available dynamic range was lost. The vivid color space setting was also making things worse by clipping the bottom end of the range, so slightly-full buckets still read as black, further reducing the available dynamic range.

The Fix

Once I understood dynamic range, and how the camera's settings were affecting it, I knew what settings to use to get the best possible range from the camera: ISO 50 and the standard color space. Armed with that knowledge, I got much better pictures at the next outdoor concert. Even so, the dynamic range was still smaller than what I'd been used to with the 2500, and I could see it in the pictures, although it was now manageable.

Lets Get Technical

So far, I've stated that the size of the pixels on the CCD matters most to dynamic range as it sets the size of the buckets. So, let's look at some numbers and illustrative pictures. Through a bit of research, I found out that the 2500 has a 1 / 2.7” CCD, which works out to pixels that are ~35.5 sq. micrometers, and the 4800 has a 1 / 2.5” CCD, which works out to pixels that are ~19.7 sq. micrometers. So here we are, the pixels on the 2500 are almost twice as large as those on the 4800. This is the explanation of why the 2500 has such better dynamic range – larger buckets. This is even evident in the images:


This is is an image of the entire scene. It included both dark shadows and very light areas, which was perfect since it should cause the metering electronics in the cameras to have to pick a middle settings that allows the lightest areas to overexpose and darkest areas to block up black. The particular region I selected to use is a tiny section of the white building and entranceway with the green awning over it, positioned over the head of the guy with the grey jacket. I selected that part because it contains both sunlit white build and dark shadow along with some midtones, all in a small region that would allow me to scale the regions large enough for people to see the lighting differences without also causing huge images.


Here is that tiny region, which includes sunlit, white-painted building and full shadow under the green awning. At the bottom of the image is the histogram from it (black on the left, white on the right). Take close note of the histogram, and especially where the bumps on the graph are, as they are the most objective illustration of the dynamic range across all the images I'll be showing.



Here's the same region from the 4800 (set to ISO50). From the image, one can clearly see the sunlit areas are completely overexposed. The histogram shows the picture far more clearly. The bumps are flatter because there are more pixels represented here and the overexposed areas are throwing off the scale (the black spike at the white end of the histogram), but the bumps are clearly shifted toward the white end of the histogram, with the dark bump pretty much where it was, but the middle bump clearly shifted toward white, and the top bump shifted right off the white end of the histogram. This is the result of the lessor dynamic range of this camera – the buckets fill faster, leading to the shifted histogram, and fill sooner, leading to the overexposed areas.

Comparison to the 5D

Now, compare the dynamic range to the 5d, which has a 35.8mm x 23.9mm sensor for a pixel size of ~67.2 sq. micrometers. One would expect it to have a better dynamic range, which is supported by this picture:



Well, first off, the focus sucks because I didn't think I'd have to look at this small of a region of the image. However, except for how soft the edges of the light and dark regions are (which will show in the histogram as wider bumps), the image and histogram both show an improved dynamic range. In this image, even the sunlit white building isn't overexposed while the shadows are not totally back either. On the histogram, not only are the three bumps from the 2500 image present (along with an extra bump in the low cluster), but they are all shifted toward the black end of the histogram, making room for a spread of all those pixels that made up the vertical bar at the white end of the histograms from the previous two cameras -- the entire image now falls between black and white without ever reaching either end. The end result is that the dynamic range of the 5D is sufficient to capture this scene entirely, even though it contains deep shadows and a full-sunlit white building.

ISO Effect on Dynamic Range

Since the 4800 and 5D have settable ISO, I also tested them at higher ISO settings. The pictures were somewhat noisier to be sure, and the ISO setting did affect the dynamic range slightly, but neither camera showed as much of an impact on the dynamic range as their inherent CCD pixel size physics, so I didn't bother showing them.

Conclusions

The dynamic range of the 5D is clearly better than either of the point-and-shoots I compared it to. That shouldn't be too much of a surprise, but it was nice to confirm photographically what the theory said, if only to give me confidence in the camera. I can also see that I'll be wanting to explore the RAW image capabilities of the camera too since I may be wanting to manipulate the images to select out the desired dynamic range for display.

posted by patbobw @ 8:54 AM