| For cameras and lenses, speed and accuracy of the auto focus are an important part of our testing procedure. We have written little about that, and we are going to change that. The differences in AF speed and AF accuracy between the various camera types, such as SLRs and mirrorless system cameras, are significant. The fastest AF is ten times as fast as the slowest! That a mirrorless system camera with contrast AF is slower than an SLR camera with phase-detection AF appears to be a fable. The lenses used do play an important role. Why is that? And what can you do with it? |
Ballet photographed in not much light, with a Canon G7, 1/1250 seconds at f/2.8. With all that running across the stage, fast AF is essential. |
How does auto focus work? |
| VFor automatic focusing, there are two basic systems: phase detection and contrast detection. We have written previously about exactly how that works (see: How does AF work?, or the articles about phase detection AF and contrast detection AF). There were also experimentation with ultrasound and infrared distance measurement, but those ultimately did not work. In addition, hybrid systems have been developed in recent years that combine the best of phase and contrast measurement.
With phase detection, the camera processor compares the image that comes from the left side of the lens with that from the right. If the image is in focus, then there is no difference between them. If it is blurred, then there are differences between the pixels that you can eliminate by shifting. From this “phase shift,” the camera can calculate how much the lens has to be turned in order to get the image in focus. Phase detection does not use the whole image for this, but certain rows of pixels, the AF fields. With simple SLRs, there might, for example, be ten of them, but for the more expensive models, you are talking about several dozen, and you can determine which field or which groups of fields you want to activate. Because you generally want to have the highest sharpness in the middle of your image, and because the sharpness for every lens is higher in the middle than in the corners, the focus fields are located around the middle of your image field. SLRs use phase-detection focusing as a standard, and then you do not focus with the shooting sensor itself (which after all only gets light as long as the mirror is folded down and pointed at the viewfinder), but with special AF sensors that are lit by half-translucent mirrors. With contrast detection, the camera searches for the image with the highest contrast: maximum focus is maximum contrast. For that, the camera uses the signal from the image sensor itself, but does not use all the pixels. Just as with phase detection, you can select certain image fields. Many compact cameras and most mirrorless system cameras (such as those from Panasonic) make use of contrast detection. An SLR camera set to Live View uses the contrast detection method. |
Contrast versus distance. We took a series of pictures of a Siemens star and adjusted the set distance by 10 centimeters. We measured the contrast with Photoshop. The vertical axis shows the contrast (standard deviation of the histogram); the horizontal axis shows the set distance. You see the maximum contrast at the focal point, distance “0”. |
| Hybrid systems have pixels on the sensor that can only look “left” or “right”. There is no separate AF sensor needed; use is made of the “regular” image sensor. With some cameras, all pixels are left or right oriented, and with others, those pixels are only placed in certain measurement fields. A left-looking or right-looking pixel of course only captures half of the light. Software corrections then have to be made. Sometimes, after focusing via the phase detection pixels, the focus is made more accurate with the contrast method. Hybrid systems have the speed of phase detection and the accuracy of contrast detection. |
Signal from body to lens: how does that work? |
| With phase detection, the body passes a signal to the lens motor: turn the lens a certain number of degrees. Expensive lenses have a feed-back loop; dthe lens has a kind of counter that checks whether the body command has been perfectly executed. Some bodies also do a second check after the setting of the lens: Is everything sharp? If so, then the picture is taken. If not, then a second round of measurements and lens rotation follows. Sometimes, the camera appears afterwards to have focused in front of or behind the area that you wanted to have sharp. This is known as the front-focus/back-focus effect.
With contrast detection, the lens always moves. When the contrast increases, the lens turns further. If it becomes lower, then it turns the other way. When the contrast no longer increases with turning, the camera concludes that focus has been achieved. Sometimes, the lens has to shift back and forth to find the right focal point. Because focusing makes use of the signal from the image sensor, you do not have problems here with the front-focus/back-focus effect: what the AF system sees is exactly the same as the final shot. Contrast detection is thus accurate. Focusing is done for any camera at full aperture. Then the focal depth is the smallest, and you thus get the most accurate measurement. Now the focus area shifts a little bit due to spherical aberrations of the lens when you stop down. That is not noticeable with less bright lenses (f/>2.8) because the focal depth simultaneously increases. You just do not notice it with very bright lenses and very precise measurements. |
Here, of course, it is about that yellow Ferrari logo here! Canon 600D with 18-135 mm, set to 100 mm, f/5.6. |
Our test method for AF speed |
| In the lab, we measure the AF speed with the camera on a tripod about 1 meter in front of the subject on which we are focusing. We measure the AF speed at full aperture, a fast shutter speed, high ISO and in AF-S mode. That means that the shot is only taken if the AF has indicated the signal as “sharp”. Because our test object always has the same brightness, the same exposure always works. We set the lens to infinity and thus go through practically the total focus range of a lens, making the differences in focus speed between different camera/lens combinations maximum. We repeat this test five times and calculate the average, the median. In fact, we measure a combination of focus delay plus release delay. Even if you let the camera focus before the shot, you still get a delay. We do not split that out any further; for most users, only the total number is relevant anyway. |
And the “winner” is…?? |
Here, of course, it is about that yellow Ferrari logo here! Canon 600D with 18-135 mm, set to 100 mm, f/5.6.
We can learn a few things from this. We found it striking that the older Nikon 60 mm macro, without a focus motor, gives nothing up in terms of AF speed to the more modern 50 mm AF-S. In general, the cameras with small sensors (Nikon CX, Four Thirds) and with internal focusing (IF) have an advantage over APS-C and Full Frame sensors. That has to do, among other things, with the much smaller lenses: the less glass there is that has to be moved, the faster. The most modern mirrorless system cameras appear to be super-fast: the two Panasonic models tested and the Nikon J5 focus faster than a Canon 1Dx. The Nikon J5 makes use of a hybrid AF system in which both phase detection and contrast detection are applied. The Panasonic GX8, the fastest camera that we have ever tested, only uses contrast detection, while that is the AF method that has a reputation of being slower (and that works with Live View on an SLR camera). The SLRs do not keep up with the mirrorless models where AF speed is concerned. |
AF speed in low light…. |
| In daylight, most auto focus systems perform fine. Few misses, although… especially when the lens has to “come from far away”, the AF of even renowned brands have difficulty finding the focal point. If the amount of environmental light decreases, that’s another story. In order to test how well the AF works with low light values, we use a “fader”. Two polarization filters that turn with respect to each other make smooth dimming to nearly total darkness possible. The elegance is that you do not have to change anything else about the test set-up. We turn the fader in a number of steps until we reach the point where the camera no longer focuses, or does so with great difficulty. We always use the same 77 mm fader; with the necessary step-up rings (here: 58 to 77) can be used on practically any lens. The camera remains at full aperture, the shutter time changes. From the shutter time and the aperture, we calculate the light value. |
Above, you can see how the AF of four cameras performs in low light. We are talking here about dim conditions, light values down to -2. First of all, it is noticeable that the mirrorless system cameras with contrast AF (Panasonic G7) are faster than an SLR camera with phase detection (Canon 760D) not only in daylight, but also in low light. Further, the Nikon D800, which is not the fastest SLR with plenty of light, appears to perform far and away the best in low light. The Nikon D800 (blue line) still focused in our test where the AF of the other three simply refused to function. We knew that the same D800 focused slowly with Live View (the red line). You also see that this body stops sooner with Live View in low light than with phase detection. |
Is the fastest also the best? |
| The temptation is great to immediately stamp the fastest AF as the best AF. That is a misconception! First, we measure the AF speed when taking a single shot. There are few cameras that are good at following a moving subject. We test continuous AF with a moving subject in practice, but we do not yet have a test set-up to be able to measure it in the lab as well. And as we showed above, there is AF speed in plenty of light and AF speed in low light. Those generally appear to differ. Some SLR camera users find the number of AF fields important. We do not share that opinion—where AF speed is concerned—since 95 out of 100 shots are made with the central focus field. A moveable AF point is mostly useful when working from a tripod or if the subject is not in the center and you do not have much focal depth.
We do think that AF accuracy is just as important as the speed. What do you gain from a fast photo if it is not perfectly focused? CameraStuffReview has of course investigated that as well. The fastest AF does not appear to automatically be the most accurate. More about that in a later article, which we hope to publish within a few weeks. Stay tuned! |