Focus Precision

As several questions have come up since the 10D announcement regarding what the 10D will or will not AF with, I thought it time to bump this thread in order to perhaps answer some of these questions.

--
Full Frame user!

EOS 3 + Nikon Coolscan III
 
I posted this in another thread as a reply, but I think it has a
wider appeal, so I am reposting it as a thread in its own right,
and making a few clarifications.

[snip]

AF systems work to a specific precision. In the canon EOS system,
there are two precisions.

Normal Precision. This is the precision that the vast, overwhelming
majority of canon EOS cameras work to. To know what cameras work to
this precision, read the list of cameras which come under the high
precision category. If it's not on there, it's normal precision.
The important ones for this forum are to note that the D30, D60 and
10D (I'm making an educated guess...you'll see why) are all normal
precision cameras.

High Precision. The only cameras capable of high-precision AF are
the EOS 1, EOS 1N (and RS), EOS 3, EOS 1V, EOS 1D and the EOS 1DS,
and HP AF is not available with all lenses, depending on maximum
aperture. With variable-aperture zoom lenses, only the slowest
aperture value is noted. With the EOS 1, HP AF is enabled when a
lens of F2.8 or faster is used. With the EOS 1N, HP AF is enabled
on the central focus point only when a lens of F2.8 or faster is
used. On the EOS 3, 1V, 1D and 1DS, HP AF is enabled on the central
focus point on an F4 or faster lens, and then a further 6 points (3
above, 3 below) when an F2.8 or faster lens is used. Note that on
all of these cameras, HP AF is combined with cross-type focus
sensors.

Canon are notably vague on the difference between normal precision
and high precision. However (and I'll check on this, but this is to
the best of my knowledge), the best definition I know of is that
Normal precision will put the lens within 100% of the DOF from the
"perfect" point of focus. High precision will put the camera within
33%. The DOF used is that of the lens wide open (so, almost
paradoxically, the smaller the DOF, the closer you will be to ideal
focus). This is one reason that faster lenses give you better
focusing (the other, of course, being that they provide the focus
sensors with more light in a given situation).

Reference: http://bobatkins.photo.net/info/faq30/eos3af.html

[ snip]

Clarifications: The EOS 10D is likely a normal precision camera
since it uses the 7 point AF system as used by the EOS 30, EOS 300
and EOS 300v, which are all normal precision cameras. The only
instance where what appeared to be the same focus system was
available in normal precision and high precision "flavours" was
that of the EOS 5 (Normal Precision) and EOS 1N (High precision as
previously described). However, the EOS 5's central AF point was
cross type to F5.6, whilst the 1N's was Cross-Type High precision
at F2.8 or faster only.

Variable Aperture Zoom lenses. Source of further confusion on the
cross-type/high-precision issue. Look at the central sensor of the
EOS 3/1V/1D/1DS AF system, and it is cross-type, high precision
(XTHP from now on) at F4 and faster. So what happens with a zoom
lens that crosses this boundary?

Let's take the popular 28-135 F3.5-5.6 IS USM as an example.
Obviously at the long end this lens can not provide XTHP focus.
However, at F3.5, the short end is fast enough for XTHP to work.
However, rather than have XTHP cut in and out, Canon decided to
disable XTHP with such a lens. Thus the 28-135 IS does NOT enable
XTHP, and neither does the 24-85 F3.5-4.5 for example. The 70-200
F4L DOES enable XTHP.

Note that with the older XTHP AF systems of the EOS 1 and EOS 1N,
an aperture of F2.8 was required, and so the 28-80 F2.8-4L USM did
NOT enable XTHP with these cameras, and was a major reason for the
demise of this lens in favour of the 28-70 F2.8L which does enable
XTHP.

--
Full Frame user!

EOS 3 + Nikon Coolscan III
--
Mishkin
 
So if I am reading this thread correctly, if a person shoots most any lens on a 10D somewhat stopped down, then your focus should fall within the depth of field which would carry the shot. Is this right?
Very nice post Isaac. I'm bumping this back up for all the 10D
focusing people to read. Wish that I'd found this sooner!

Yeang
 
--Good post...thanks. I hope that the ones doing the ff and bf
tests read it. It might make them feel better in knowing why the
focus will not be perfect.
It is a good post. Unfortunately, I'm finding rear focus outside of 100% of the DOF, thus it isn't meeting normal precision standards :)

Simon
 
The DOF used is that of the lens wide open (so, almost
paradoxically, the smaller the DOF, the closer you will be to ideal
focus). This is one reason that faster lenses give you better
focusing (the other, of course, being that they provide the focus
sensors with more light in a given situation).
Practically, it all boils down to having more light to work with.
Not quite so. I applied more thought to this.

Take two lenses of equal focal length. Let's say the 135 F2L and 28-135 F3.5-5.6 IS at 135. Let us say that we're at a subject distance of 3m.

Now, take a normal precision camera body. AF will put the lens within 100% of the DOF (calculated for Circle of Confusion of 0.035mm) at the wide open aperture of the lens. So, for the 135 F2L the focus is within 6cm (since the DOF at F2 at 2m is 2.97 to 3.03m according to http://www.canon.com/camera-museum/camera/lens/ef/data/ef_135_2l_usm_dof.html ) of "perfect" focus, regardless of what aperture we then stop the lens down to.

If we now look at the 28-135 IS, the focus of the lens falls anywhere within almost 30cm of "perfect" focus (DOF at 135 F5.6 3m is 2.869 to 3.144m, http://www.canon.com/camera-museum/camera/lens/ef/data/ef_28~135_35is_usm_dof.html )

If we now stop both lenses down to F8, the 135 F2L is still focused MUCH more precisely on the subject, with it falling much more closely in the centre of the DOF, resulting in a sharper image.

If we now mount these lenses on a 1DS, the 135 F2L presses its advantage even further, since it enables high-precision focus, and the lens is now focused to within 2cm of "perfect" focus, whilst the 28-135 remains within 30cm.

--
Mostly Full Frame user!

EOS Tree + Nikon Coolscan III
Deef Hurty.
 
It is a good post. Unfortunately, I'm finding rear focus outside
of 100% of the DOF, thus it isn't meeting normal precision
standards :)
But are you?

How are you deciding that your image is out of focus? Bear in mind that canon use a Circle of Confusion of 0.035mm for all EOS system DOF/AF. The pixels on a 10D or D60 are 0.0074mm in size, so the DOF for single pixel sharpness is very much less than the DOF defined by the COC.

If you're looking at the output 100% on a monitor, that is an unrealistic enlargement. Think about it. 1280x960 on a 19" monitor viewing a D60/10D image at 100% is around a 39X enlargement from a 22.5x15mm sensor. How would it look making a 30"x20" print from a piece of film that size? Because that's what you're doing looking at it at 100%, and asking for single-pixel sharpness viewed from 12" away is deeply unrealistic.

--
Mostly Full Frame user!

EOS Tree + Nikon Coolscan III
Deef Hurty.
 
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html
Just a thought, but those CoC/DOF calculations are based presumably on a full frame sensor. I'm not sure, but with the 1.6x crop factor, we are scrutinizing an image with a 60% greater degree of magnification I presume.

Thus what is an acceptable CoC figure gets trashed by the 1.6x effective magnification of the image? In order to get an equivalent CoC equivalence on the 10D (or acceptable zone of focus), would any DOF calculation not have to be based on a CoC figure 1.6x smaller? Could this be at the root of the perceived problems?

Simon
 
It is a good post. Unfortunately, I'm finding rear focus outside
of 100% of the DOF, thus it isn't meeting normal precision
standards :)
But are you?
I'm not sure... :) See below....
How are you deciding that your image is out of focus? Bear in mind
that canon use a Circle of Confusion of 0.035mm for all EOS system
DOF/AF. The pixels on a 10D or D60 are 0.0074mm in size, so the DOF
for single pixel sharpness is very much less than the DOF defined
by the COC.
I'm not looking at single pixel sharpness actually. The measure I was trying to use was to see where the point of camera focus actually was in relation to the point of desired focus, then trying to see with a DOF calculator, if that point of sharp focus was within the zone provided by the calculator. ie, a relative measurement not based on how sharp I perceived a pixel to be.
If you're looking at the output 100% on a monitor, that is an
unrealistic enlargement. Think about it. 1280x960 on a 19" monitor
viewing a D60/10D image at 100% is around a 39X enlargement from a
22.5x15mm sensor. How would it look making a 30"x20" print from a
piece of film that size? Because that's what you're doing looking
at it at 100%, and asking for single-pixel sharpness viewed from
12" away is deeply unrealistic.
Yes, I know - see above. That's absolutely not what I was doing.

Simon
 
100% of DOF normal-precision focusing on 10D/D60 means that if the camera focused just on the edge of DOF field, the lens will project a 0.035mm CoC on sensor, thus covering 5 pixels instead of one.

That can't be right. I rarely see such blurry pictures (each pixels blurred onto 5). So is that 100% DOF figure for NP AF right?
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html

--
Mostly Full Frame user!

EOS Tree + Nikon Coolscan III
Deef Hurty.
--
Mishkin
 
only 20% of all autofocused shots would be pixel-sharp. That doesn't correspond to reality.
That can't be right. I rarely see such blurry pictures (each pixels
blurred onto 5). So is that 100% DOF figure for NP AF right?
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html

--
Mostly Full Frame user!

EOS Tree + Nikon Coolscan III
Deef Hurty.
--
Mishkin
--
Mishkin
 
How are you deciding that your image is out of focus? Bear in mind
that canon use a Circle of Confusion of 0.035mm for all EOS system
DOF/AF. The pixels on a 10D or D60 are 0.0074mm in size, so the DOF
for single pixel sharpness is very much less than the DOF defined
by the COC.
If Canon considers 7.4u-pixel sensor covered with 35u CoC acceptable, that's simply ridiculous. That means, pixels are terribly wasted. 22.7x15.1mm effectively has VGA resolution when covered with 0.035mm spots (22.7/0.035=649, 15.1/0.035=431, 649x431=0.28 megapixels. Ouch!). Whny then have 6.3MP on that sensor? Just in case that accidentally, with 20% chance the CoC due to focus [in]accuracy will be less than 7.4u?
If you're looking at the output 100% on a monitor, that is an
unrealistic enlargement. Think about it. 1280x960 on a 19" monitor
viewing a D60/10D image at 100% is around a 39X enlargement from a
22.5x15mm sensor. How would it look making a 30"x20" print from a
piece of film that size? Because that's what you're doing looking
at it at 100%, and asking for single-pixel sharpness viewed from
12" away is deeply unrealistic.
No matter what, each and every of 6.3 million pixels must record valuable information, must be sharp. Otherwise, it doesn't make sense to have 6MP in the first place. Pixels these days are still larger than physical limits posed by lens aberrations and diffraction, so not having each pixel record every bit of resolution the lens is capable of is lame.

Canon must wake up and redesign their AF system so that maximum CoC due to AF inaccuracy is less than pixel size.

--
Mishkin
 
I have been thinking for a while on exactly this question - I am glad you asked.

The issue is whether the 1.6 cropping factor affects the CoC. If it is a "cropping" factor it would seem that it should have no effect.

Anyone that can explain why one way or the other - please enlighten me.

(PS - Isaac, thank you for such an informative thread! Refreshing.)

Frank
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html
Just a thought, but those CoC/DOF calculations are based presumably
on a full frame sensor. I'm not sure, but with the 1.6x crop
factor, we are scrutinizing an image with a 60% greater degree of
magnification I presume.

Thus what is an acceptable CoC figure gets trashed by the 1.6x
effective magnification of the image? In order to get an
equivalent CoC equivalence on the 10D (or acceptable zone of
focus), would any DOF calculation not have to be based on a CoC
figure 1.6x smaller? Could this be at the root of the perceived
problems?

Simon
--
I plan on living forever - so far so good!
 
Interesting thought but I wonder how this applies to the 24-85. This lens was introduced with the APS SLR and theoretically should have had a smaller CoC. If this is the case, one would expect that this lens would in fact be more accurate. On the 10D, it is one of the worst performers in terms of focusing accuracy.
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html
Just a thought, but those CoC/DOF calculations are based presumably
on a full frame sensor. I'm not sure, but with the 1.6x crop
factor, we are scrutinizing an image with a 60% greater degree of
magnification I presume.

Thus what is an acceptable CoC figure gets trashed by the 1.6x
effective magnification of the image? In order to get an
equivalent CoC equivalence on the 10D (or acceptable zone of
focus), would any DOF calculation not have to be based on a CoC
figure 1.6x smaller? Could this be at the root of the perceived
problems?

Simon
 
The crop factor has no relation to DOF.

Just think you shoot with full frame and crop the part equivalent to the sensor size of 10D.
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html
Just a thought, but those CoC/DOF calculations are based presumably
on a full frame sensor. I'm not sure, but with the 1.6x crop
factor, we are scrutinizing an image with a 60% greater degree of
magnification I presume.

Thus what is an acceptable CoC figure gets trashed by the 1.6x
effective magnification of the image? In order to get an
equivalent CoC equivalence on the 10D (or acceptable zone of
focus), would any DOF calculation not have to be based on a CoC
figure 1.6x smaller? Could this be at the root of the perceived
problems?

Simon
--
Go ahead, never look back
 
with a D60 or 10D the COC is much smaller. My program would calculate it as something between 19 and 27 dependiing on the desired print size and lines/mm of the print.
Canon use a Circle of Confusion of 0.035mm in all DOF calculations
for the EOS system. DOF charts for ALL EF lenses may be found here:
http://www.canon.com/camera-museum/camera/lens/f_lens.html

--
Mostly Full Frame user!

EOS Tree + Nikon Coolscan III
Deef Hurty.
--
Bob Poole
 
The crop factor has no relation to DOF.

Just think you shoot with full frame and crop the part equivalent
to the sensor size of 10D.
No, that's not what I'm saying - or at least, that's not what I mean. The DoF calculators for a given focal length state a zone of acceptable focus. However, in our cameras a 100mm lens (whilst still having the dof of a 100mm lens) produces an image that is effectively a 1.6x crop.

The images we produce with our 1.6x crop factor are basically 60% enlargements for a given print size. Thus what would be acceptible DOF based on CoC, has been zoomed in (for a given image size) by 1.6.

Can you see what I mean?

Simon
 
I was thinking the same thing.

Saying a "crop" has the same DoF is to ignore that one has to enlarge that crop more than the un-cropped image, to get a given print size.

At the greater enlargement, fewer parts of the image will be within the CoC limits set as "acceptable", ...i.e. DoF.

Sure like to see a good explanation of why this is not so.

Larry (BTW, Sincere Thanks to Isaac for the info presented throughout this thread! :-)
The crop factor has no relation to DOF.
Just think you shoot with full frame and crop the part equivalent
to the sensor size of 10D.
No, that's not what I'm saying - or at least, that's not what I
mean. The DoF calculators for a given focal length state a zone of
acceptable focus. However, in our cameras a 100mm lens (whilst
still having the dof of a 100mm lens) produces an image that is
effectively a 1.6x crop.

The images we produce with our 1.6x crop factor are basically 60%
enlargements for a given print size. Thus what would be acceptible
DOF based on CoC, has been zoomed in (for a given image size) by
1.6.

Can you see what I mean?

Simon
 

Keyboard shortcuts

Back
Top