Busted! Digital Photography Myths
Photography seems to have more than its fair share of erroneous beliefs and misunderstandings which get reinforced by repetition without question. Here's a collection of some, often seen in the Canon forums, debunked.
If there's one simple message you can take from this document, it's try it yourself! Don't look for other people to define the boundaries of your photography, do a little research and experiment with your camera to find out what's true for you.
The Canon EF 50mm f/1.8 II Has a Crude Focus Mechanism
Many Canon shooters have been frustrated by inconsistent autofocus (AF) with the EF 50mm f/1.8 II lens (the "50/1.8"). It sounds like a home-made agricultural implement when it is autofocusing, and the manual focus ring feels gritty and has a lot of backlash, so it seems obvious that any AF problems must be due to the apparently crude mechanical nature of the lens. But is that the reason?
I have a 50/1.8 that was purchased new from B&H in 2008. It was calibrated with the EOS 450D I had at the time by the Canon Australia service centre in Sydney in 2009. That combination focused perfectly with contrast detection (CD) AF in Live View, perfectly some of the time with phase detection (PD) AF through the viewfinder, and in the same place in front of the subject the rest of the time with PD AF through the viewfinder. That front-focused (FF) place put the subject outside the depth of field (DOF) at f/2.8 near the minimum focus distance (MFD). One shot in perfect focus, the next clearly out of focus in the same place, so...
Is it the camera?
On an EOS 60D, the calibrated 50/1.8 does the same. It gives perfect focus using CD AF –
|Image 1: calibrated lens, 60D, CD AF, in focus (click on the image to see it full-size)|
If I take a bunch of shots with PD AF using the centre AF point, with the lens out of focus each time before triggering AF, some of them will be in perfect focus –
|Image 2: calibrated lens, 60D, PD AF, in focus (click on the image to see it full-size)|
And the rest of them will show the subject in the same place clearly behind the DOF –
|Image 3: calibrated lens, 60D, PD AF, front focus (click on the image to see it full-size)|
The same tests give the same pattern with an EOS 7D, and a similar in-focus/back-focus pattern with an EOS 5D, so the random-focus-in-two-places behaviour is not an anomaly of any one of the four Canon DSLRs tested so far. That suggests it's the lens, so...
Is it a particular lens, or is it characteristic of the model?
Another 50/1.8 purchased in 2013 but not calibrated gives the same pattern. On a 60D, the uncalibrated 50/1.8 gives perfect focus every time using CD AF –
|Image 4: uncalibrated lens, 60D, CD AF, in focus (click on the image to see it full-size)|
And a slight front-focus compared to CD AF and the calibrated lens some of the time with PD AF using the centre AF point –
|Image 5: uncalibrated lens, 60D, PD AF, in focus (click on the image to see it full-size)|
And the plane of focus is clearly in the same place in front of the subject the rest of the time with PD AF using the centre AF point –
|Image 6: uncalibrated lens, 60D, PD AF, front focus (click on the image to see it full-size)|
(All of the above are typical shots from sets of ten to twenty. Before you condemn the tests for using a slant target, check out You Can't Test Autofocus with a Slanted Target. I used f/2.8 for all test shots because it gives a sharp enough image to be able to evaluate where focus is, whereas at much wider apertures the softness of the image makes that difficult.)
The same pattern appears with the uncalibrated lens on a 7D or 5D – randomly in focus or out of focus the same amount, e.g. about 7mm for both lenses on the day they were tested with the 60D and 7D, as shown in images 5 and 6 above. (It wasn't entirely random with my calibrated 450D. If I started with the lens front-focussed it would consistently confirm the in-focus point, but since none of the other bodies I've tested do that, and in normal shooting the lens will be randomly front or back-focussed when you trigger AF, I'll keep calling it random.) So the reality we are dealing with is PD AF focuses the 50/1.8 randomly in two distinct places, and the safe bet at this stage is it seems to be characteristic of that lens in general. Whether one of those places will be in good focus or not depends on the state of calibration of the body and lens in question.
There are a few 50/1.8 owners who say their lens is not typical, that it's sharp at f/2 and always focuses well. I have no reason to disbelieve them, but I only have evidence from the two lenses I have tested, one of which has been through a Canon service centre calibration process and the other which is only a year old, so I have to consider that those two instances are typical, especially since their behaviour is so consistent with each other. This article relates to that typical behaviour.
The "crude mechanism" myth
The popular explanation for this random-focus-in-two-places behaviour is that the "focus mechanism" (the motor and/or mechanical components and/or controlling electronics) of the 50/1.8 are "crude", that the lens can only position focus by coarse increments. In other words, if position X is in focus, then positions X+1 and X-1 are out of focus and can't be brought any closer to each other.
For instance, lets say position X of the focus mechanism puts focus 500mm from the image sensor, then position X+1 puts focus at 507mm and X-1 at 493mm, therefore a target at 500mm will either be in focus or 7mm (or 14mm...) out of focus. You would expect if you moved the camera 4mm and refocussed, the target would end up either 3 or 4mm out of focus.
But it never is. No matter which camera body I use, which individual 50/1.8, on any target, on any day, in any decent light, whether I pre-focus in-front of or behind the target before triggering AF, and, no matter how far I move the camera between two shots, all shots are either in focus (depending on calibration) or the same distance out of focus in the same direction (generally front-focussed). I never get some front-focussed and the rest back-focussed as the "crude mechanism" myth predicts. I have done that test with camera movements from 0.1mm to beyond 7mm, and no "crude" shift in focus has ever been observed. Busted!
You will get the same two points in both One-shot and AI Servo. You can also find them by setting the lens to manual focus (MF) mode, and moving the camera towards and away from the target while holding a half-press on the shutter button, or repeatedly half-pressing, and listening for the focus confirmation beep. I use a macro slide with a 1.25mm screw pitch and an Allen key in the longitudinal shaft, which means I can move the camera with some precision down to about 0.05mm. There is no randomness in that case (the direction of approach determines which point you get - the farthest if you start with front-focus and the nearest if you start with back-focus), and focus will be confirmed throughout the 7mm range between the two points. That means the two points are determined by what the AF sensor sees, not by how the focus mechanism moves. See Busted! The Myth of Open-loop Phase-detection Autofocus for further proof that it's all about what the AF sensor sees.
How precise is the focus mechanism?
CD is the most consistent AF method we have, so if there are crude steps in the focus motion of the 50/1.8 we should be able to see them by moving the camera a small distance between each shot and observing how the focus changes. The "crude mechanism" myth predicts that focus will move with the camera then suddenly shift when the next focus position comes into effect.
Here's a video of 50 shots shown at four frames per second - download the static video (WMV, 950KB). There was no camera or target movement between the shots, and the camera was refocussed in Live mode before each shot. This video shows hows consistent CD AF is from shot to shot when there's no movement.
Here's a similar video with the camera moved 0.1mm closer to the target and refocussed before each shot - download the moving video (WMV, 860KB). To remove the change in the size of the image of the target as the camera got closer, the frames were stacked in Photoshop with the "Attempt to Automatically Align Source Images" option so they don't jump around relative to each other and they look the same apart from where the focus is.
Can you see any significant difference in accuracy of focus between the two videos? For instance a jump or saw-tooth pattern in the second one as the camera moves through 5mm towards the target? If not, it means the smallest increment of focus motion the 50/1.8 can take is something less than 0.1mm of focus distance. That's 5% of the thinnest possible DOF according to standard calculations, or 0.02% of the MFD. Busted again!
When PD AF focuses in one of two places it does so with only a tiny bit less precision than CD AF does with its single point (as found when running the tests which result in the six images above). The accuracy is about the same but it's slightly less repeatable, with a few more outliers. So there's no evidence that PD AF is not able to use the finest resolution of focus movement that the lens is capable of. The reality we are dealing with is PD AF randomly focuses the 50/1.8 precisely in two distinct places, determined by what the AF sensor sees.
The bottom line is, the "crude" myth predicts that focus occurs in two places because the mechanism in the lens is not able to resolve any finer movement. But when given precise instructions by CD AF the mechanism is able to position focus to better than 0.1mm of object space near MFD. It's only when PD AF is telling the lens where to put focus that the two points of focus appear, and focus is achieved at each of the two points with only a tiny bit less precision than the single point produced by CD AF. Totally busted!
But why DOES it focus in two places?
This question always comes up – if it's not a crude mechanism in the lens, then why does the typical 50/1.8 randomly focus in two consistently different places with PD AF? I'm happy to tell you what I know, but just a reminder that this article is about proving that the 50/1.8 doesn't have a crude focus mechanism, it's not about explaining what does cause the random focus-in-two-places behaviour.
First a little background information... CD AF removes the problem by finding the point of focus which gives the highest contrast on the image sensor. PD AF takes light from an annulus just inside the f/5.6 entrance pupil for "standard precision" AF sensors, and another just inside the f/2.8 entrance pupil for "high precision" AF sensors. A PD AF sensor looks at light from two sides of the lens, through the appropriate annulus, and tries to bring the two images of the subject into alignment by changing the focus (it's a closed-loop control system, not "one measurement, one movement", see Busted! The Myth of Open-loop Phase-detection Autofocus).
The best theory I've come across to explain the focus-in-two-places problem is that light coming through the f/2.8 annulus is affected enough by spherical aberration that it makes it difficult for the AF sensor to determine where the two images come into alignment. In other words, the light from there is not perfectly clear and unambiguous, as shown in the first diagram on the Wikipedia article about Spherical Aberration. Specifically, it's proposed that spherical aberration leads to asymmetrical bokeh (the appearance of out-of-focus areas is different in front of and behind the plane of focus), which confounds the AF sensor.
The spherical aberration theory is supported by a couple of experimental results. F/5.6 AF points also give focus in two places, but closer together than you get with f/2.8 points, as you'd expect since the light they see (from closer to the lens axis) should be less affected by spherical aberration or asymmetrical bokeh.
I tried putting an external aperture (a hole in a card) on the front of the calibrated 50/1.8, to block light entering the outer annulus of the objective element, which should reduce the amount of aberrant light reaching the AF sensor. I used a diameter that gave the same shutter duration in Av with the lens set to f/1.8 as the naked lens gave at f/2.8, hoping that would let in the same amount of light as f/2.8 while not obstructing the view through the lens of the f/2.8 AF sensor. With an external aperture of that diameter and the lens set to f/1.8, the two points of focus are significantly closer together again.
Since it's proven not to be a mechanical and/or electronic problem due to crudeness within the lens, it seems likely to be an optical problem. The distance between the two points of focus reduces when we give the PD AF sensors light that should be less affected by spherical aberration, which does suggest it's an optical phenomenon, and seems to support the theory of spherical aberration as the primary cause. In other words, the problem seems to be caused by the PD AF sensor's inability to resolve an unambiguous alignment of images, due to the asymmetrical bokeh inherent in the optical design of the 50/1.8. But that's for someone else to prove or disprove. Not being able to prove why doesn't take away the fact that the typical 50/1.8 does focus precisely and randomly in two places with PD AF.
Always Turn Off Your Camera When You Change Lenses
On Canon Europe's Infobank Sensor Cleaning page, it says, "You cannot stop dust getting into your digital camera, but you can reduce the risk a little... Switch the camera off before changing the lens. This reduces the static charge on the sensor and stops it attracting dust." It goes on to recommend minimising the time your camera is open, and avoiding dusty conditions. Sounds like good advice, for "a little" risk reduction.
For this to take on the status of a commandment, we would have to accept the following:
- the sensor is charged when the camera is on,
- the charge is strong enough to attract dust from outside the camera,
- even though the sensor is behind an anti-aliasing filter with an air gap between the two,
- even though the AA filter is behind the closed shutters, and
- when the camera is switched off, the charge is reduced sufficiently to make a difference.
According to Canon guru Chuck Westfall, "Canon CMOS image sensors do not carry electrostatic charges at any time." (Tech Tips, April 2007)
I don't know if the rest of those points are valid, but if we must turn off our cameras to minimise dust entry, why doesn't Canon tell us to do that in the instruction manuals for our cameras and lenses?
Don't Use AI Servo (continuous autofocus) with Static Subjects
In Rudy Winston's highly recommended A Look at The Canon Autofocus System series of videos, he says:
"If you put the camera in [AI] servo and point it at something that is totally stationary, the system can get a little freaked-out, and it can start – it's assuming there's gotta be movement, there's gotta be movement, there's gotta be movement – and if it sees a total absence of movement it can actually start jittering the lens kinda back and forth looking, you know, feverishly for this movement that it assumes must be there." (part 1 of 3, 17:11-17:32)
So let's test it. The critical thing is to make sure that the camera and the subject are absolutely motionless relative to each other, so that we aren't observing the system responding as it should to changes in the camera-to-subject distance. That means a close fixed target, camera on a rock-solid support, single autofocus point, image stabilisation turned off, and a remote release to trigger autofocus (preferably wireless so that no disturbance is transmitted from your hand). The best way to judge whether focus is stable or changing is to listen with your ear close to the lens.
The result with numerous lenses on several bodies (including the very first EOS from 1987), is no evident jitter whatsoever. If you find a combination that does jitter under those conditions, please let me know.
So why do people think AI Servo will jitter with a static subject? Because when they try it hand-held, the system actually starts jittering the lens kinda back and forth responding, you know, feverishly to this movement they are causing by not being able to hold the camera perfectly still. They experience the system responding exactly as it should to the changes in the camera-to-subject distance that they are causing, but they think the system is getting a little freaked-out. See Busted! The Myth of Open-loop Phase-detection Autofocus to understand why phase-detection autofocus systems don't jitter when the camera-to-subject distance is unchanging.
Canon EOS Cameras Have a Fill-flash Mode
It's often said that Canon EOS cameras fire the flash at a lower output level in aperture priority mode (Av) to lighten shadows ("fill flash"), and at a higher output level in manual mode (M) or program mode (P) to illuminate the whole scene.
It's easy to disprove this:
- Set the metering mode to centre-weighted average and flash metering mode to average to rule out any fancy metering tricks. Don't use Live View because that forces evaluative metering.
- Set up a shot in Av with a shutter duration like 1/60 or 1/125.
- Match the aperture, shutter, and ISO settings in M.
- Verify that both modes give the same image brightness.
- Deploy the built-in flash and take a shot in both modes.
- Is the M shot brighter than the Av shot?
- Try it at various combinations of exposure compensation (EC) and flash exposure compensation (FEC). (If nothing changes when you use positive FEC, the flash is firing at full output already, so bump up the ISO to give it some headroom.)
If the myth were true, at least some of the M shots would be brighter than their equivalent Av shots but none of them are, so we know that the flash metering is independent of the shooting mode.
So why do people think that flash is brighter in M? The other side of the mode-independent flash story is that Canon also makes the ambient metering flash-independent (with one exception). In other words, it (generally) doesn't make any difference to the settings you get in Av mode with the flash on or off. So (generally) when you use flash in Av you get the same image brightness from the ambient light, and you notice the contribution of the flash in the shadows – that's fill flash. In M it's easy to reduce the contribution of the ambient light so that the flash provides most of the illumination, but the flash is firing with the same output as it would with the same settings in Av.
The exception is Negative Evaluative Exposure Compensation (NEVEC), which uses a faster shutter in Av, or a tighter aperture in Tv, to reduce the ambient exposure in low light. It only happens in evaluative ambient metering mode, and doesn't seem to affect P mode.
Perhaps there is confusion with the Auto Fill Reduction feature, which seems to operate in evaluative flash metering mode by automatically applying negative flash exposure compensation in bright light. In that case the behaviour is a part of evaluative flash metering and is not dependent on shooting mode.
That document (Flash Photography with Canon EOS Cameras - Part II) doesn't explicitly define "fill flash mode", but it seems they mean, "the camera... always tries to expose the background adequately". That's exactly what normally happens in Av and Tv, so at least in those shooting modes "fill flash mode" is not a mode at all, it's business as usual, same as "no flash mode". There does not appear to be a corresponding "not-fill flash mode". It also says, "Canon EOS cameras always default to fill flash mode when the camera is in Tv, Av and M modes", which makes perfect sense, since if you centre the meter in M you will get the same ambient and flash brightness as in Av with zero compensation.
In P mode, shutter duration is constrained to be no slower than 1/60 of a second (so it's effectively a manual exposure), which might cause the background to be underexposed and therefore make the flash dominate. Similarly in M mode you can choose any shutter duration you want, so you are in complete control of the balance between ambient and flash illumination.
When the scene is dimly lit, Av will give you a long shutter, and P will give you a lot of flash, but that's not because the camera switches into "fill flash mode", it's because in both cases the flash metering adjusts the output to illuminate the scene. When the scene is brightly lit, the flash meter sees that and adjusts the output accordingly. In both cases that's just flash metering doing it's own job regardless of shooting mode. P makes it complicated by effectively giving you a manual exposure but showing zero on the meter as if 1/60 were "right". That change in shutter duration has no influence on the flash metering, which just does what it does to give a normally bright image.
Ambient metering, flash metering, and shooting mode are all fundamentally independent of each other, despite some features that may alter certain things for their own purposes. Fill flash is not a mode, it's what happens when ambient light dominates the image brightness and flash output is metered accordingly. You can do that in any shooting mode. See the Flash section of the unofficial Rebel Forum FAQ for some simple guidance on using flash with EOS cameras.
Phase-detection Autofocus is an Open-loop Control System
This one seems to be particular to the Canon DSLR world, due to the famous Canon autofocus information post by RDKirk on the Fred Miranda forums, backed up by dubious statements by Canon guru Chuck Westfall.
The simple truth is, phase-detection autofocus confirms focus when the autofocus sensor sees an in-focus subject. See Busted! The Myth of Open-loop Phase-detection Autofocus to prove it for yourself.
You Will Break Your Lens If You Turn the Focus Ring in AF Mode
This may be a Canon-only myth as well. In their instruction manuals for lenses which don't have the Full-time Manual focus (FTM) feature, Canon say things like:
- "Do not touch the rotating parts of the lens while autofocus is active." (EF-S 18-55mm IS, EF-S 55-250mm IS)
- "Do not touch the focusing ring on the lens during autofocusing." (EF50mm F/1.8 II)
- "Do not adjust focus manually when the focus mode switch is set to AF." (EF-S 18-55mm IS, EF-S 55-250mm IS)
(FTM lenses are designed so that you can change the focus after autofocus is confirmed, and this myth doesn't have anything to do with those lenses.)
If we take these statements as warnings, what could they mean? Surely if they mean that you will break your new lens if you do anything like the above, Canon would say so in big red capitals all over the manuals, and deliver new lenses with warning labels stuck around the focus rings. If it were true, I would expect such a warning to read more like, do not touch the focusing ring on the lens with the focus mode switch set to AF, as this will damage the focus mechanism and we won't fix it under warranty!
What if we take these statements as advice? Don't interfere with the focus ring while the lens is autofocusing... because that's going to make it slower, and might cause inconsistent behaviour or errors? Sure.
Don't manually focus in AF mode... because in that mode it's hard to turn the ring accurately and smoothly, and when you press the shutter button later the focus will change, so you're much better off to switch to MF mode? Definitely.
Of course you don't want to stress and wear the focus gears in your lens by winding the focus ring forcefully back and forth in AF mode. And you definitely don't want to overpower the focus motor while it's working. Nobody would advocate doing anything like that. But that's not what the myth is about.
One day I was playing around with a 450D and 55-250mm lens in AF mode. I put the camera down on a table. I didn't particularly think about or pay attention to the fact that the focus ring was touching the table and supporting the weight of the lens. While I was doing something else the camera went to sleep, so when I came back I touched the shutter button to wake it up. The camera swung itself across the table as the focus ring acted like a wheel on an axle! If the myth were true, surely that would have broken the lens, but it has gone on to reliably focus many thousands of times, and continues to behave perfectly normally years later.
Inevitably, ignorant photographers are going to turn their focus rings deliberately to pre-focus in AF mode, and innocent photographers are going to turn their focus rings by accident when they don't check the focus mode. But the myth requires us to believe that Canon can get away with selling products which are guaranteed to fail in real-world use by ordinarily fallible users.
How else could a focus ring get moved while the lens is inactive in AF mode? While putting on a lens cap, hood, or filter, the photographer might inadvertently touch the focus ring. Force on the end of the lens might cause the focus ring to turn (e.g. the EF-S 18-55mm IS does this). While mounting or unmounting the lens, a stronger grip and gross movement is required, so it's likely to happen then as well. A focus ring could get touched or brush against something in normal handling, such as grasping the camera by the lens or putting it into a camera bag.
Multiply those probabilities by the millions of "vulnerable" lenses in the world and how often they are used, and by any estimate you have a significant number of perilous events every day. If all you had to do to break your lens was to touch the focus ring, the Canon Rebel Talk and Canon SLR Lens Talk forums would be overrun with I touched my lens and it broke!! threads. And yet there is not a single credible report anywhere I can find of a lens ever being harmed that way.
When asked to produce evidence of lens failure in this way, believers usually come up with the kit lens -18-55 auto-focus broken :( discussion on Flickr. Is this a credible report?
A not credible report sounds like, My lens stopped AFing (no specific cause). I wonder if the ribbon broke? (unverified speculation about why it stopped autofocusing). Maybe I made things worse by turning the focus ring while it was in AF mode? (unverified speculation about what might have caused it, probably based on the myth).
A credible report would sound like, My lens was working perfectly in AF mode. I turned the focus ring by hand and heard/felt a crunch/snap. Immediately it stopped AFing. I sent it to Canon and they replaced the focus gears. Here's a copy of the repair report (image), "Focus gears replaced due to mechanical failure." Here's a picture of the broken gears (image). The technician said, "don't ever touch the focus ring in AF mode because it damages the gears." If you find a credible report anywhere, please let me know.
So what should you do with your non-FTM lenses?
- Don't touch the focus ring while autofocus is active with the lens in AF mode, in other words, while autofocus is happening, while the focus confirmation light is lit or blinking, or while the shutter button or back button is pressed, either directly or in effect by remote control.
- Do switch to MF when you want to focus manually, even if you use the back button to trigger autofocus.
- Don't worry that you will break your lens if the focus ring gets turned while the autofocus system is not active.
You Can't Do Exposure Compensation in Manual Mode
Whenever someone says something like, I dialled in -1 exposure compensation by changing the shutter speed in manual mode, someone will immediately respond with something like,You can't do exposure compensation in manual mode!?! You can only do it in semi-automatic modes, like aperture priority!
(Unless otherwise stated, "manual mode" means full manual with fixed ISO, in this document. Manual with auto-ISO is a distinct semi-automatic mode, like aperture priority, shutter priority, and program.)
The short answer is, you need a camera function to compensate the exposure in a semi-auto mode, but you don't need a function to compensate the exposure in manual mode, and you're doing exposure compensation either way. If that makes sense to you there's no need to read on.
Who's right? Both photographers are, according to their different definitions of "exposure compensation". The first definition is broad and embraces the second, and the second is narrow and incorrectly denies the first (it's really more of a delusion than a myth).
Usually, the second photographer has come to photography in the digital era, and has never seen the term "exposure compensation" in any context other than in the name of a function in the instruction manual for his digital camera. (The heading for that section in the instruction manual may be something like "Setting Exposure Compensation", rather than just "Exposure Compensation", which tells you that exposure compensation is a separate thing in its own right to how it is set.) Instruction manuals are written to tell you how to make the camera do what you already know you want to do, they aren't intended to give you a grounding in the fundamental concepts of photography and why you would want to use those concepts, so they predominantly describe how to do it, not what it is and why you do it.
Digital camera instruction manuals describe how to "compensate" (bias) the "exposure" (image brightness) with a dedicated function of the camera, which allows you to apply a fixed offset to the default metering in semi-auto modes. (Some cameras, including almost all Canons, don't allow you to do that in manual with auto-ISO mode, even though it is a semi-auto mode and there's no reason why you shouldn't be able to.) The offset is automatically applied by the exposure meter, so that the setting or settings it chooses make the images darker ("underexposed") or brighter ("overexposed") compared to what you would get without the offset (with the meter on zero). The exposure meter display typically shows the amount of the offset in Exposure Value (EV) steps. So the instruction manuals for digital cameras only describe a function which provides automatic compensation (using a fixed offset to vary the settings), and they use "exposure compensation" as a short label for that function. It's like, even though "post-mortem" is used as shorthand for "post-mortem examination", it doesn't make "post-mortem" only mean autopsy, it still literally means "after death".
Usually, the first photographer learnt the basics of photography in the film era and learnt about exposure compensation as a general photographic concept. She knows that it means using different settings to what the exposure meter wants, to get darker or brighter images. This definition includes the second as a particular case of how to do it. She knows that it doesn't matter to the image brightness how you do it, as long as the "exposure" (the image brightness that results from the settings used to take the shot), gets "compensated" (biased, offset, adjusted, corrected, deviated... from what zero on the meter would give). There is nothing about the word "compensation" that restricts its meaning to the automatic application of a nominal amount of bias, or that a bias can only apply to the exposure settings (aperture and shutter speed), and processing settings (ISO) that you give the meter the task of choosing.
The first photographer probably had a film camera with a dedicated mechanical dial to set compensation, and the instruction manual for that camera probably called it something like an "exposure compensation dial" (e.g. Pentax 67II, see page 54). The dial simply offset the zero point on the meter, affecting all metering and modes the same. It did that by simply changing the ISO setting, so that the meter operated as if she were using a faster or slower film. Or she could do that herself by deliberately setting the "wrong" ISO for the film she was using.
In a semi-automatic mode (e.g. aperture priority), the meter would select a value for the dependent setting (shutter speed) so that the meter needle indicated zero, and the exposure was compensated by the amount of the offset. In manual mode, if she used settings that put the meter needle on zero, her exposure was again compensated by the amount of the offset. If she used settings which did not zero the meter, the needle indicated how far she was from where the fixed offset put zero, so the total compensation was the sum of the dial offset and the meter reading. In manual mode she was doing manual compensation (using fixed settings to get a variable deviation).
By the way, "exposure compensation" in the photographic (what it is) sense is an obsolete term in the digital era. In the olden days you could only change the luminous exposure (aperture x shutter duration x scene luminance) with the camera settings, but now that we can change the image processing (ISO) shot-to-shot, the "exposure" in "exposure compensation" is anachronistic. It's better to think of "metering compensation" or "image brightness compensation". Similarly for "underexposed" and "overexposed"... "underbright" and "overbright". (See Exposure vs. Brightening.)
When the first photographer uses a digital camera, she recognises that it does things differently – the exposure meter display typically shows the deviation of the current settings from zero, the same in every mode. In a semi-auto mode (e.g. manual with auto-ISO), the deviation is the offset she specified using the function described in the instruction manual. In full manual mode the deviation is how far her settings are from what the meter says will give default image brightness. In other words, the exposure meter display is simply a mode-independent image brightness compensation meter.
The simulator at Exposure Compensation: Your new best friend shows this perfectly. Note that the discussion below the simulator is written according to the second definition, but it's clear that the author understands that it's where the meter needle is, rather than how it got there, that matters for the image brightness.
You can easily demonstrate this for yourself:
- Set up your camera on a tripod with a static scene in unchanging light.
- In shutter priority mode with a fixed ISO, set the compensation to zero, and take a shot. Say you got 1/100 and f/8.
- Go to full manual mode, use the same settings (including the ISO), and take a shot. Same image brightness, and the meter display also shows zero, right?
- Back in shutter priority mode, dial in -1 compensation, and take a shot. It should be at 1/100 and f/11. The image looks one stop darker than in step 2, right?
- Back in full manual mode, change the aperture so that the meter display shows -1, and take a shot. Same settings and image brightness as in step 4, right?
In both modes, the "exposure" (image brightness) was "compensated" (adjusted) the same amount (-1, as shown by the meter display) in the same way (closed the aperture one stop), and the image brightness changed correspondingly and consistently. Depending on how your camera works, it might also be true that you touched the same user interface controls in the same way (maybe the opposite direction), to make the same adjustment.
That is the essence of the first photographer's definition. It accommodates how compensation is done differently in the two modes. In shutter priority mode, the adjustment was relative to zero on the meter – the meter used -1 to automatically give her f/11. In full manual mode, the manual adjustment was absolute – the meter showed her that f/11 was -1 from zero.
Is that a valid definition? How is "exposure compensation" defined outside of digital camera instruction manuals? Here are a few definitions that Google found:
- "a technique for adjusting the exposure indicated by a photographic exposure meter, in consideration of factors that may cause the indicated exposure to result in a less-than-optimal image."
- "the act of overriding a camera's automatic exposure in order to achieve a particular effect or due to difficult lighting conditions."
- "Deliberately changing the exposure settings recommended by a light meter in order to obtain a different exposure to better fit personal preferences, create special effects or meet special requirements."
- "Lighten or darken the image by overriding the exposure system."
- "Modifying the shutter speed and/or lens aperture recommended by the camera's light meter in order to produce special creative effects or to meet special requirements."
- "increasing or decreasing the exposure to compensate for dark or light subjects which may cause a light meter to indicate the 'wrong' exposure."
None of those definitions explicitly restrict the idea to an automatic function of the camera in semi-auto modes, although you could say that the second one might be suggesting that. The trick is to differentiate between definitions of what it is from descriptions of how to do it.
I don't know when the term was first used, but it is certainly recognised as a general concept well before the digital era, in film camera instruction manuals and literature about photographic technique. For instance, "exposure compensation" appears in volume 21 of American Photography magazine with no introduction, which shows that the author did not need to explain the term in 1927. Formulas and tables of compensation values for standard situations, like using bellows and filters, were published in the '30s and '40s. Ansel Adams' Zone System is acknowledged as a specific form of exposure compensation, which was developed by the late 1930s and is founded on work from the late 19th century. Adam's did not have an exposure compensation function on his camera.
Whether it was called "exposure compensation" or not, the concept pre-dates the 20th century. For instance, an 1898 essay on Alpine Photography discusses "the exposures required... at an altitude of about 6000 feet", compared to what gives "a satisfactory negative of an open English landscape, on a bright June day with fleecy clouds in the sky" (-2 stops, according to the author).
Is "exposure compensation" ever used to describe metering in full manual mode in contemporary digital camera instruction manuals? One example is the Sony A850. The Sony instruction manual isn't clear, but David Busch's guide explains "how exposure compensation works when shooting with Manual exposure mode on the A850" (page 126). It's the digital equivalent of the exposure compensation dial on a film camera like the Pentax 67II, simply offsetting the metering by the amount required.
Another example is the Panasonic DMC-FZ200. The Operating Instructions for advanced features describes the camera's "Auto Bracket" function as applying an "exposure compensation range" in semi-auto and full manual modes (page 146). This works the same as the auto bracketing function of any other typical digital camera, for instance, with the meter on +1 and an "exposure compensation range" of -/+1 EV, you get shots at +1, 0, and +2. As is typical of most camera instruction manuals, the function is only described in terms of how to do it, and a broader definition of what and why isn't given.
Consider auto bracketing in general. The second definition of "exposure compensation" could be stated as, a function to automatically apply a fixed offset to the default image brightness settings to get a darker or brighter image. A typical bracketing function automatically applies two fixed offsets (e.g. -1 and +1) to get one default, one darker, and one brighter image. You can do that in semi-auto modes (even in manual with auto-ISO on a Canon), and you can do it in full manual mode. The default is wherever the meter needle happens to be at the time, either by a fixed relative offset from zero in a semi-auto mode, or by a variable offset according to the fixed absolute settings and scene luminance in full manual mode. From my point of view, that means you can do exposure compensation in full manual mode, according to the second (how to do it) definition, thus invalidating any claim that you can't.
The second photographer is probably thinking, but that's not called "exposure compensation" in the camera manual, so it can't be exposure compensation. To be able to claim that his definition is the only right one, he would have to prove that "exposure compensation" has never been used to refer to anything other than a function of digital cameras, which is disproved above.
The second photographer will also claim that no-one uses the term in the general sense these days. It's not hard to find examples:
- Ken Rockwell is clearly using the broad definition when he talks about Perfect Exposures with Large Format Cameras (not a general endorsement of Rockwell, just an example of someone currently using the original definition).
- Klaus Schroiff uses "compensation" and "correction" synonymously in the photographic sense on photozone.
- A thread about Exposure compensation in Photoshop? in the Canon EOS 7D / 10D - 70D Talk forum shows that no-one involved in that discussion raised an eyebrow about the idea of doing exposure compensation in the digital darkroom in 2003.
- An informal survey of participants in a 2013 discussion of Exposure Compensation in the same forum found that 45% were comfortable with the traditional definition, and would have no trouble understanding what the first photographer means by using it.
- Ferrell McCollough wrote a section on "Manual Exposure Compensation" in his 2008 book Complete Guide to High Dynamic Range Digital Photography.
- Jim Doty, Jr wrote a section on "Using exposure compensation with the meter in manual mode" in his 2011 book Digital Photography Exposure For Dummies.
- How to calculate the exposure compensation needed when using extension tubes or bellows describes "why adding extension makes exposure compensation necessary".
The essential thing to grasp is that an exposure meter just balances four factors, aperture, shutter speed, ISO, and compensation. You can set values for up to three of them, and it will choose values for the rest depending on the luminance of the scene. For example, for aperture priority you set aperture, ISO, and compensation, and it chooses the shutter speed it thinks will make the image brightness "correct". For shutter priority with auto-ISO, you set shutter speed and compensation, and it gives you a combination of aperture and ISO (according to the logic it knows to work that out). For full manual, you set aperture, shutter speed, and ISO, and it gives you compensation. As far as the meter is concerned, the four factors are just numbers in a calculation, it makes no difference which are designated as input and output via the mode selection, and the various displays on the camera body typically show all four values the same in every mode. You just pick a mode that suits which factor(s) you want to be changed for you as the luminance changes.
Which definition is right? It's certainly valid to say in any company or context, I compensated the image brightness -1 by changing the shutter speed (and/or aperture, and/or ISO) in full manual mode. (The first photographer would think, yeah, exposure compensation...) It's also perfectly obvious that you generally can't use the function labelled "setting exposure compensation" or "exposure compensation" in your digital camera's instruction manual, to compensate the image brightness in full manual mode (though there is at least one exception to that). You generally can do relative compensation in full manual mode on a digital camera via the automatic bracketing function, even though the instruction manual will generally not use the term "exposure compensation" to describe that function. It's ridiculous to say that you can't compensate the image brightness in full manual mode, since the only time you're not doing that is when the meter display happens to show zero.
A simple way to avoid confusion is to explicitly state which form of compensation you are doing. For instance, I'm using the exposure compensation function (so you must be doing it automatically with a fixed offset in a semi-auto mode), I'm doing exposure compensation in aperture priority (obviously you're using the automatic function to do it), I'm manually compensating the exposure by fixing the settings (so you must be in full manual mode), or, I compensated the exposure in full manual mode (obviously you're doing it manually with fixed settings). It's generally not hard to work out from the context what's going on. Type-one photographers need to keep in mind that type-two photographers get very confused when manual or absolute compensation is mentioned, because they are adamant that "compensation" can only mean the name of a camera function to do automatic relative compensation.
The bottom line is:
- When you use settings which make the exposure meter display a non-zero value, you are compensating the image brightness.
- That is how "exposure compensation" has always been defined.
- There are two types of compensation – relative to what the meter determines the image brightness should be, and absolute.
- It's possible to do relative compensation by getting the camera to automatically apply a fixed offset in semi-auto modes (the exposure meter determines settings to match the offset), and in manual mode (the exposure meter offsets its readings).
- It's possible to do absolute compensation by using fixed settings in full manual mode (the exposure meter determines the offset which results from the manual settings).
- It's also possible to do relative compensation by getting the camera to automatically apply a fixed -/+ offset range in both semi-auto and full manual modes. Bracketing works in combination with any other relative/automatic or absolute/manual compensation which may be in effect.
- All ways to do it have the same effect on image brightness for the same amount of compensation.
- The amount of compensation will typically be shown on all meter displays the same in all modes.
In other words, a typical digital camera's exposure meter display is simply and always a mode-independent image brightness compensation meter.
You Can't Test Autofocus with a Slanted Target
Slanted focus targets are often used for cheap, simple, or improvised autofocus (AF) accuracy testing. See Jeffrey’s Autofocus Test Chart for a general discussion. Many claim that all slanted targets are invalid, for instance, citing Canon's standard autofocus test method, which specifies a target parallel to the image sensor at a distance of at least fifty times the focal length. See What is the best way to use the Micro Focus Adjustment on the Mark III? in Chuck Westfall's December 2008 Tech Tips at The Digital Journalist.
Below a specification of the method, Westfall advises:
"Do not attempt to autofocus on an angled chart, because doing so will degrade the consistency of the camera's focusing measurement. Keep in mind that the camera's AF sensor is comprised of multiple pairs of linear pixel arrays. If you attempt to autofocus on a single line in an angled focusing chart, only a few pixels from each active pixel array will 'see' the target. Ideally, the contrast in the reference target should cover the entire area of the camera's centre focusing point, and the reference target should be perfectly parallel to the camera's focal plane."
His meaning is not entirely unambiguous. Does "a single line in an angled focusing chart" mean a single line of many on an angled chart, or an angled chart on which there is a single line? The single line of many case (e.g. a rule) is perfectly clear – it won't work, since the AF sensor can't tell which of the multiple lines it can see is the one you want to focus on.
A good slanted target, like Tim Jackson's (alternative download), Jeffrey Friedl's, or Yvon Bourque's isn't like that. It presents a single line perpendicular to the lens axis, and therefore the only thing the autofocus sensor can see is "perfectly parallel to the camera's focal plane", when the target is used correctly. But the contrast provided by such a target does not cover the entire length of the AF arrays.
Such targets are designed to be used near the minimum focus distance (MFD). They are not suitable for use with the Canon method, which specifies a target beyond fifty times the focal length (50f) for the purpose of autofocus calibration with reference equipment. Some argue that if Canon say the target must be beyond 50f then testing at anywhere closer is invalid. Are our cameras not meant to focus close to MFD? Close-up, within "point blank range"? Not even macro lenses? It may be that calibration at a longer distance is the better compromise for typical use of tricky lenses, such as extremely wide-aperture primes that don't focus consistently over their focus range.
My Canon 450D showed consistent autofocus errors using Jackson's target. I sent it with the three lenses I had at the time to the Canon Service Centre in Sydney, New South Wales. They agreed it had an error and calibrated everything under warranty. Here are a couple of images from the calibration process that they left on the memory card for me (full images, resized and sharpened):
|Image 1: Canon autofocus test - EOS 450D, EF-S 18-55mm f/3.5-5.6 IS at 18mm and f/5.6|
|Image 2: Canon autofocus test - EOS 450D, EF-S 18-55mm f/3.5-5.6 IS at 55mm and f/6.3|
After calibration, all three lenses focussed perfectly on Jackson's target all the way down to their MFDs. In other words, at least for those three lenses on that camera, if Jackson's target shows an error near MFD so will Canon's at a longer distance, and if Canon's target shows no error so will Jackson's. This will not be true for all lenses, for instance the Canon EF 50mm f/1.2L is notorious for focussing differently depending on subject distance.
From the look of it, I guess Canon's target is about the size of a business card, or maybe a small smart phone. If so, those two images were shot from close to one metre away, not the 2.75m minimum that Canon's method specifies for a 55mm lens. If the images were shot from 2.75m, the target would have to be more like 200mm wide. So it seems that Canon's 50f rule is one they don't mind breaking. If you know of a valid technical reason why Canon specify 50f, beyond "they say so", please let me know.
Coming back to what "a single line in an angled focusing chart" means, Leonard Shepherd says Chuck Westfall e-mailed him this advice in 2006:
"I recommend using a flat, detailed target parallel to the focal plane. After reading through the PDF linked from your message [Tim Jackson's target], it appears that the author has missed a major point, i.e., any individual focusing point in a digital SLR is much longer than the simple line he is using on his chart. The nature of the AF sensors used by EOS digital SLR’s as well as those from other manufacturers is that they perform most reliably when the entire length of the focusing area sees readable detail. This condition is not satisfied by a thin line on a piece of paper. It's OK to include an angled chart in a test photo. In fact, Canon Factory Service Centers always do this [as shown above]. But the test target is always separate from the angled chart, and parallel to the camera's focal plane." (See message 124624 at Focusing charts for D300/D3).
So it's clear that Westfall did mean an angled chart with a single line on it. It should be easy to show that such a target "degrades consistency" and is not the "most reliable", compared to an ideal parallel target, which provides contrast to the full length of the AF sensor.
I have compared the speed, consistency, and accuracy of multi-line and single-line parallel targets, and found no difference in good light with a Canon EOS 60D and a Canon EOS 650 (the original EOS from 1987, with a single horizontal AF sensor), close to the MFD.
Differences start to appear in low light. The 60D doesn't appear to have any problem at levels above the AF-assist threshold. Close to the 60D's threshold, the 650 confirms focus on a parallel multi-line target but fails on a single line. Below the threshold, it takes a few more AF-assist flashes (and therefore a slightly longer time), for the 60D to confirm focus on a single line. I don't see any difference in accuracy if focus is confirmed, only in speed. So it's true that a target with sensor-spanning detail is optimal, but a thin target is no problem in good light.
How thick does a single line have to be to work as a parallel focus target? Tests with the Canon EF 100mm f/2.8 L lens on the 60D at 50f (5m) show that a single stroke from a 0.4mm fibre-tip writing pen works fine. In a full-resolution image, that line is 1 to 2 pixels wide. Yes, you read that right, the 60D can focus on a line that is only two image pixels wide. Chuck Westfall said it's not ideal if, "only a few pixels from each active pixel array will 'see' the target". I agree for challenging conditions, but with current gear in good light it doesn't matter enough to prohibit all slant targets on that basis.
If you are able to find a way to show a significant difference in speed, accuracy, or reliability between multi-line and single-line parallel targets in good light (e.g. bright open shade or interior illumination), please let me know.
But tests with a single line angled target at 50f do show that arrangement is not accurate or reliable, and the depth of field is so deep that a small 45° scale is poor at showing the position of the plane of focus. So Canon are right to say don't use a slanted target at distances like 50f.
I have extensively tested Tim Jackson's target with phase-detection (PD) and contrast-detection (CD) autofocus (AF) on various Canon, Nikon, and Sony DSLRs and found it to be as accurate and reliable as any parallel target when used as intended (near MFD in good light). If you find a way to show a problem under those conditions, please let me know.
The following images show the performance of the centre PD AF point of a Canon EOS 60D with an EF 40mm f/2.8 STM lens (f/2.8, 1/30, ISO 100), with Jackson's single-line target at various angles:
At what angle do you judge that autofocus performance is obviously degraded? When does the target start to appear to be far enough from the centre of the depth of field that you would make the error of thinking that the system requires adjustment?
Those images are not cherry-picked, they are essentially a series of one shot at each angle. Multiple shots at any particular angle showed consistent and reliable focus.
You can see a touch of back-focus in the slanted images. I checked this using a perpendicular target with a 45° scale on the side (made from Jackson's target). Compared with CD shots, PD shots show the same amount of back-focus. In other words, the back-focus you see in the above shots is due to the performance of PD AF with that body and lens combination, not due to the slant of the target.
At angles below 15°, AF starts to take noticeably longer (with the individual steps in the closed-loop process becoming apparent), but I was able to get reliable focus confirmation down to 7° with the standard Jackson target. Below that, it was apparent that the blur from other elements in the image was starting to merge with the target line.
How low can you go with a target that's clear above and below the "Focus here" line? At 3°, focus is still confirmed quickly and accurately:
|Image 9: 3°, PD AF|
|Image 10: Detail of 3°, PD AF|
At 2°, focus is consistently achieved, but is not confirmed:
|Image 11: 2°, PD AF|
|Image 12: Detail of 2°, PD AF|
Contrast-detection AF runs into a limit around 10°, where it fails in an unmagnified Live view but easily achieves and confirms focus in a 5x magnified Live view:
|Image 13: 10°, contrast-detection autofocus at 5x magnification|
Canon's CD AF only works on vertical detail. Simple slanted targets are criticised for only showing horizontal detail to a camera in landscape orientation, so therefore they don't work that way with CD AF, and only test the vertical (or diagonal) elements of phase-detection autofocus sensors. That could also be an advantage if it suits your testing purpose.
In the case of Jackson's target, vertical detail is provided by the "Focus here" text, when used close to MFD. Tests with CD AF and a single horizontal PD AF point show that it gives consistent accurate autofocus in both modes. Use the camera in portrait orientation to test CD AF at greater distances, or the horizontal elements of PD AF sensors near MFD, with a simple slanted target.
I guess that Canon's slanted scale (Images 1 and 2) is somewhere around 15° to the lens axis. That makes sense since at subject distances like 50f you need a long scale at a low angle to give a clear indication of focus error, and in that case you must use a parallel target. So Canon's prohibition of slanted targets fundamentally derives from using targets at distances like 50f. It's also true that if you want to test autofocus near MFD the necessity for a shallow scale and a parallel target goes away, and as shown above that works perfectly for that purpose.
The bottom line is:
- Don't ever use a slanted target with more detail up or down the slope than a single line (perpendicular to the lens axis) on a blank background.
- There's no problem with a parallel or slanted single-line target near MFD with recent cameras in good light.
- Don't use a single-line target with an old camera in low light.
- On parallel targets in low light, PD AF sensors benefit from more detail than a single line on a blank background, but in good light it doesn't matter.
- Canon's specification of a parallel target is only necessary for target distances significantly beyond MFD, like 50f. That also requires a low-angle scale, as shown in images 1 and 2.
- Slanted single-line targets are consistent, reliable, and accurate when used as intended, i.e. in good light near MFD.
In other words, if you want to test at 50f then follow all of the rest of Canon's advice and use a parallel target with a low-angle scale. If you want to test near MFD you can use a good slant target with complete confidence. There's no reason we know why you shouldn't test near MFD, only that it's not the way Canon say they do it. Do we have to follow Canon's rules if they don't?
To be included when I get around to it:
Hyperfocal Shooting Makes Everything In Focus
Dark and Light Scenes Fool/Trick the Meter
Raising the ISO Makes the Sensor More Sensitive
You Should Put a Filter on Your Lens for Protection
You Should Learn Composition With a Fixed Focal Length Lens
An Image Straight Out of the Camera is Authentic and Post-processing is Cheating
Wide-aperture Lenses Allow More Light to Reach the AF Sensor
If you're aware of one that isn't included, or an aspect that isn't covered, please let me know.