Metering

Metering in modern digital cameras has become an incredibly complex undertaking, and the metering system in the 5D mark III is no exception to that. The core of the iFCL (intelligent Focus, Color, and Luminance) metering system is a 63-zone “dual-layer” sensor similar to the one introduced in the EOS 7D.
As Canon says on the matter:

Electronic sensors in general are more sensitive to red light. This means when photographing subjects with lots of red in them – skin tones for example – the sensor receives a stronger signal as it only detects brightness levels. This can lead to the wrong assumption that there is more light than there really is.
The Dual-layer system overcomes this by having one layer sensitive to red/green light and one layer sensitive to blue/green light. Both these layers measure the light in their respective spectra and the metering algorithm then combines the two to provide an accurate light reading. In this way, accurate exposures can be attained in a wide range of shooting situations and irrespective of the colour of the subject being metered.

However, it’s important to remember that when Canon is talking about knowing the color of the subject being metered, they aren’t talking about what most reviewers seem to interpret that to mean. The color capabilities of the iFCL dual layer meter are significantly restricted compared to what most reviewers seem to expect, and honestly, I would be very surprised if Canon used the provided color information for anything more than simply addressing the stated over-sensitivity to red wavelengths and the color temperature adjustments in the AF system that they claim.

I was curious as to whether the color weighting that Canon has included worked as well as they suggest it should, so I under took a fairly simplistic, and certainly not laboratory grade, test to determine the accuracy of the meter.

For this tests, I used a Canon EF 70-200mm f/4L IS USM focused at infinity, and set the camera to an aperture of f/4, ISO 100, a white balance to 6500K (to match the D65 white balance of my display, not that the color temp should affect the metering), and spot and evaluative metering. Since the rest of the exposure data remained constant, I’ll just be reporting the metered shutter speed. I used 3 colors, 255:0:0 red, 0:0:255 blue, and 128:128:128 gray for the tests, displayed on my profiled NEC Spectraview II display.

Now before I get to the numbers, a bit of reasoning here. Infinity focus was chosen because at the distances where I can fill the frame with a 24” display, the cameras will actually resolve the individual pixels, and I wanted to insure that the meter wasn’t affected by that. Further, I expect some variation in metered values for each of the colors. The simple reality is I don’t have a perfectly accurate way to insure that the display is producing the same power for each of the colors. Because of that, what’s important is the difference between the metered value in live view and the metered value with the iFCL meter, not the values metered by the two cameras separately.

The first test, which I considered something of a control, used my 1D mark III to produce a set of sample images in both live view and using the standard single-layer meter. These results are summarized in the table below.

The same process was repeated with the 5D mark III. The results of which are summarized in the table below.

In looking at the results, I’m somewhat disappointed in the performance of the 5D mark III. With reds, the iFLC meter appeared to be spot on, matching the imaging sensor’s reading exactly. However, in the other two cases, the meter tends to underexpose compared to the metering in live view. The average deviation for both cameras was the same 0.56 stops, but the standard deviation is much tighter at 0.19 stops instead of 0.51 stops for the EOS-1D mark III. Both spot and evaluative metering performed nearly identically on both cameras.

In my opinion, a slight underexposure is preferable when metering with the separate metering sensor. Since digital capture clips at saturation and loses data after that point, having the camera back off slightly when working with values from a metering sensor that’s not identical to the imaging sensor seems somewhat appropriate.

However, the uniformity of the readings on the 5D mark III was disappointing. The EOS-1D mark III’s meter is within 1/3 of a stop for each of the targets, while the 5D mark III’s varies from matching to a stop under exposed. This seems to be precisely the behavior that Canon claims the dual layer sensor is supposed to compensate for.

That said, this wasn’t the most scientific experiment, and there’s certainly enough room for error in the screen and the rounding of the camera’s meter that the difference is probably somewhat less than it appears. As it stands, I’m not quite sure what conclusion I’m ready to draw from this. My 5D mark III definitely underexposed the blue and gray screens in the test, but I don’t really see that same behavior in real world situations.

As interesting as the sensor hardware might be, the user facing side of the system isn’t appreciably different from any of Canon’s previous cameras. The metering system operates in the same 4 modes as it has in the past, in order of decreasing coverage, evaluative, center-weighted average, partial, and spot.

Spot metering on the 5D mark 3 covers approximately 1.5% of the visible image. Unlike the 1D bodies, the 5D mark 3 doesn’t have multi-spot metering capabilities, nor can the spot meter be linked to the active AF point. The area covered by the spot meter is indicated in the viewfinder by the dashed circle in the center.

Partial expands the sport metering coverage to approximate 6.2% of the viewfinder. Partial metering is not actually indicated in the viewfinder, which again seems like an omission given that Canon draws all of the viewfinder lines in the viewfinder dynamically as part of the LCD driven intelligent viewfinder.

Center weighted average is the old standby from the days of yore. Unlike evaluative, center weighted average doesn’t attempt to throw sophisticated models and technology at the problem it simply weights the segments near the center of the frame more than the segments towards the edges.

Evaluative metering is where all the magic in the Canon system takes place. Under ideal circumstances, the evaluative algorithms will integrate data from a number of sources, when available, including:

• Exposure and color data from the dual layer-metering sensor.
• Focus data, namely which points are in or near focus, from the AF system.
• And focus distance from distance capable lenses.

All the fancy hardware aside, the real question is how good does it all work?

In practice, it certainly seems like it does. In general, I always shoot in evaluative unless I’m testing something specific and shoot in spot. Even in strongly backlit scenarios, and scenes with radically changing conditions from strong front lighting to back lighting, I’ve never really been let down by the evaluative meter.

I’ve made a few unsuccessful attempts to look at the meter’s performance in more detail since I got the camera, but lacked sufficient tools to be able to do anything definitive. Instead, I’ve mostly looked at the output of general shooting and I have little to nothing to complain about in the way the 5D mark III meters.

Canon specifies the metering range as being EV 1-20 with an EF 50mm f/1.4USM at ISO 100. In my testing back when the camera was first released, and people were complaining about a light leak affecting the meter, I found the meter, at least in my camera, worked as expected over the specified EV 1-20 range.

Then again, maybe I’m just damning the meter with faint praise here.

Coming from a 1D mark III, the iFCL meter in the 5D mark III doesn’t knock my socks off. Its range is more limited, EV 1-20 instead of EV 0-23; it doesn’t introduce more zones; and the color data doesn’t seem to make the meter readings any more accurate in practice.

Compared to the EOS-1D X, or even the camera’s Nikon contemporaries, both of which use approximately 100,000 pixel RGB sensors, it’s hard not to call the meter lacking—at least on paper. The color data is a step in the right direction. However, the lack of color accuracy and spatial density, means that the camera can’t do face detection at all, and any color based AF tracking is likely extremely limited if it can even do color based AF tracking at all.

In some ways, I really feel like the meter in the 5D mark III is yet another example of the conservative stance large camera manufacturers take to product development. It’s a step forward, but it’s a step forward at a glacial pace. Then again, the dual layer sensor does mean that Canon has real world experience in producing and using stacked pixel photo sensors, which ultimately may have implications in the imaging sensors in the future.