# Re: Megapixel Limit

Discussion in 'Digital Photography' started by Kennedy McEwen, Jun 23, 2011.

1. ### Kennedy McEwenGuest

In article <%fGMp.24571\$>, Gary
Eickmeier <> writes
>Serious question - based on rumor of Sony a77 camera with 24 MP on an
>APS - C chip - is there a practical limit to megapixels, beyond which
>no lens can resolve any more detail?

Diffraction.
Radius of the smallest blur is 1.22·L·f/#, where L is the wavelength of
light. Roughly speaking, the closest that two points can be resolved is
where the peak of the blur on one is coincident with the edge of the
blur on the other - the Rayleigh Criterion. So there is little point in
having a pixel smaller than half the diffraction blur radius - that
gives you one pixel on the peaks and one in the trough between them. You
can go a little further because the Rayleigh Criterion still has 8%
modulation, but you need infinitesimally small pixels with exceedingly
low fill factors to discern that. At L·f/# there is zero modulation
from the lens. Nevertheless, with reasonable pixel fill factors, the
Rayleigh Criterion represents the practical limit, with the smallest
pixels half that size. After that its just how big do you want to make
your sensor - and how much will you pay for it!
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Kennedy McEwen, Jun 23, 2011

2. ### Kennedy McEwenGuest

In article <>, Alan Browne
<> writes
>On 2011-06-23 13:15 , Kennedy McEwen wrote:
>> In article <%fGMp.24571\$>, Gary
>> Eickmeier <> writes
>>> Serious question - based on rumor of Sony a77 camera with 24 MP on an
>>> APS - C chip - is there a practical limit to megapixels, beyond which
>>> no lens can resolve any more detail?

>>
>> Diffraction.
>> Radius of the smallest blur is 1.22·L·f/#, where L is the wavelength of
>> light. Roughly speaking, the closest that two points can be resolved is
>> where the peak of the blur on one is coincident with the edge of the
>> blur on the other - the Rayleigh Criterion. So there is little point in
>> having a pixel smaller than half the diffraction blur radius - that
>> gives you one pixel on the peaks and one in the trough between them. You
>> can go a little further because the Rayleigh Criterion still has 8%
>> modulation, but you need infinitesimally small pixels with exceedingly
>> low fill factors to discern that. At L·f/# there is zero modulation from
>> the lens. Nevertheless, with reasonable pixel fill factors, the Rayleigh
>> Criterion represents the practical limit, with the smallest pixels half
>> that size. After that its just how big do you want to make your sensor -
>> and how much will you pay for it!

>
>for a C-sized sensor of 25 Mpix (about: 6120 x 4080 pixels over
>24*16mm) the max sensor diameter per pixel is about 3.9 uM.
>
>Per your equation, green is about 1.8 uM when shot with a f/1.8 lens
>
>With that, the sensor in question should be able to resolve detail even
>with the actual senor site being a little smaller than above.
>
>I did that quickly in a spreadsheet, so please point out any errors I
>

The main error is that few, if any, f/1.8 lenses deliver diffraction
limited resolution when used wide open. They need to be closed by at
least a couple of stops to reach their sweet spot. So put f/4 in your
have already exceeded the capability of the optics they sit behind. ;-)

Optical improvement isn't as sexy as sensor improvement and there isn't
a single figure that salesmen (they aren't marketing!) can use to sell
it.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Kennedy McEwen, Jun 24, 2011

3. ### Kennedy McEwenGuest

In article <>, Alan Browne
<> writes
>On 2011-06-23 20:07 , Kennedy McEwen wrote:
>> In article <>, Alan Browne
>> <> writes
>>> On 2011-06-23 13:15 , Kennedy McEwen wrote:
>>>> In article <%fGMp.24571\$>, Gary
>>>> Eickmeier <> writes
>>>>> Serious question - based on rumor of Sony a77 camera with 24 MP on an
>>>>> APS - C chip - is there a practical limit to megapixels, beyond which
>>>>> no lens can resolve any more detail?
>>>>
>>>> Diffraction.
>>>> Radius of the smallest blur is 1.22·L·f/#, where L is the wavelength of
>>>> light. Roughly speaking, the closest that two points can be resolved is
>>>> where the peak of the blur on one is coincident with the edge of the
>>>> blur on the other - the Rayleigh Criterion. So there is little point in
>>>> having a pixel smaller than half the diffraction blur radius - that
>>>> gives you one pixel on the peaks and one in the trough between them. You
>>>> can go a little further because the Rayleigh Criterion still has 8%
>>>> modulation, but you need infinitesimally small pixels with exceedingly
>>>> low fill factors to discern that. At L·f/# there is zero modulation from
>>>> the lens. Nevertheless, with reasonable pixel fill factors, the Rayleigh
>>>> Criterion represents the practical limit, with the smallest pixels half
>>>> that size. After that its just how big do you want to make your sensor -
>>>> and how much will you pay for it!
>>>
>>> for a C-sized sensor of 25 Mpix (about: 6120 x 4080 pixels over
>>> 24*16mm) the max sensor diameter per pixel is about 3.9 uM.
>>>
>>> Per your equation, green is about 1.8 uM when shot with a f/1.8 lens
>>>
>>> With that, the sensor in question should be able to resolve detail
>>> even with the actual senor site being a little smaller than above.
>>>
>>> I did that quickly in a spreadsheet, so please point out any errors I
>>>

>> The main error is that few, if any, f/1.8 lenses deliver diffraction
>> limited resolution when used wide open. They need to be closed by at
>> least a couple of stops to reach their sweet spot. So put f/4 in your
>> spreadsheet and you'll get an answer that indicates that most sensors
>> have already exceeded the capability of the optics they sit behind. ;-)
>>
>> Optical improvement isn't as sexy as sensor improvement and there isn't
>> a single figure that salesmen (they aren't marketing!) can use to sell it.

>
>I doubt my 135 f/1.8 has such issues.

Which 135mm F/1.8 lens would that be?

>To be conservative, I could plonk f/2.5 in there and it would still be
>fine.

Well, just for the sake of demonstration, here's the Carl Zeiss Sonnar-T
135mm f/1.8, which itself is no shoddy performer, on full 35mm format.

Pull the slider on the left from f/22 down to f/1.8 to see the effect of
diffraction on resolution. It improves down to about f/8, but doesn't
actually get any better going down to f/1.8 (certainly not in proportion
to the f/# as expected for just diffraction) and even degrades at lowest
f/#s. Even though some parts of the image, such as the centre, remain
as sharp at f/1.8 they aren't 4.5x sharper. So, for this 135mm f/1.8
the effective diffraction limited f/# is around f/8.

>Indeed, per the spreadsheet, the limit appears to be "effective" f/5.6
>more or less - I'd expect a decent quality lens of f/2.8 to just make
>the criteria.
>

Few do.

>My zoom lenses, albeit high quality, would probably be at or beyond limit.
>

I doubt it, but tell us what they are - the internet is littered with
these type of resolution reviews, so I am sure we can find them.

>What it also means, at least for really good lenses, is that there is a
>more room for FF sensors to grow in resolution.

You'll notice I said "few", and there are indeed a few that perform to
the diffraction limit wider than f/5.6. I doubt that there are any that
perform to the diffraction limit wide open, even your 135mm f/1.8, and
certainly no zooms.

For an example of a zoom, try the legendary Canon 70-200 f/2.8L IS II:
m
Significant flattening out of the blur after reaching its diffraction
limit at f/8 - f/5.6 throughout the focal length range. The Nikon
equivalent is similar.

So, yes, there is more room for more megapixels on FF, but it will only
be realised with an extremely small number of prime lenses. Most
manufacturers will need vast improvements to their optics range to make
such increase in sensor megapixels worth it. That won't stop them
increasing megapixels though - there are far more customers who believe
"more is better regardless" than there are customers who know it isn't
or will even try to check.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Kennedy McEwen, Jun 27, 2011
4. ### Kennedy McEwenGuest

In article <>, Alan Browne
<> writes
>On 2011-06-27 09:20 , Kennedy McEwen wrote:
>> In article <>, Alan Browne
>> <> writes
>>> On 2011-06-23 20:07 , Kennedy McEwen wrote:
>>>> In article <>, Alan Browne
>>>> <> writes
>>>>> On 2011-06-23 13:15 , Kennedy McEwen wrote:
>>>>>> In article <%fGMp.24571\$>, Gary
>>>>>> Eickmeier <> writes
>>>>>>> Serious question - based on rumor of Sony a77 camera with 24 MP on an
>>>>>>> APS - C chip - is there a practical limit to megapixels, beyond which
>>>>>>> no lens can resolve any more detail?
>>>>>>
>>>>>> Diffraction.
>>>>>> Radius of the smallest blur is 1.22·L·f/#, where L is the
>>>>>> wavelength of
>>>>>> light. Roughly speaking, the closest that two points can be
>>>>>> resolved is
>>>>>> where the peak of the blur on one is coincident with the edge of the
>>>>>> blur on the other - the Rayleigh Criterion. So there is little
>>>>>> point in
>>>>>> having a pixel smaller than half the diffraction blur radius - that
>>>>>> gives you one pixel on the peaks and one in the trough between
>>>>>> them. You
>>>>>> can go a little further because the Rayleigh Criterion still has 8%
>>>>>> modulation, but you need infinitesimally small pixels with exceedingly
>>>>>> low fill factors to discern that. At L·f/# there is zero modulation
>>>>>> from
>>>>>> the lens. Nevertheless, with reasonable pixel fill factors, the
>>>>>> Rayleigh
>>>>>> Criterion represents the practical limit, with the smallest pixels
>>>>>> half
>>>>>> that size. After that its just how big do you want to make your
>>>>>> sensor -
>>>>>> and how much will you pay for it!
>>>>>
>>>>> for a C-sized sensor of 25 Mpix (about: 6120 x 4080 pixels over
>>>>> 24*16mm) the max sensor diameter per pixel is about 3.9 uM.
>>>>>
>>>>> Per your equation, green is about 1.8 uM when shot with a f/1.8 lens
>>>>>
>>>>> With that, the sensor in question should be able to resolve detail
>>>>> even with the actual senor site being a little smaller than above.
>>>>>
>>>>> I did that quickly in a spreadsheet, so please point out any errors I
>>>>>
>>>> The main error is that few, if any, f/1.8 lenses deliver diffraction
>>>> limited resolution when used wide open. They need to be closed by at
>>>> least a couple of stops to reach their sweet spot. So put f/4 in your
>>>> spreadsheet and you'll get an answer that indicates that most sensors
>>>> have already exceeded the capability of the optics they sit behind. ;-)
>>>>
>>>> Optical improvement isn't as sexy as sensor improvement and there isn't
>>>> a single figure that salesmen (they aren't marketing!) can use to
>>>> sell it.
>>>
>>> I doubt my 135 f/1.8 has such issues.

>>
>> Which 135mm F/1.8 lens would that be?

>
>Sony 135 f/1.8.
>
>>
>>> To be conservative, I could plonk f/2.5 in there and it would still be
>>> fine.

>>
>> Well, just for the sake of demonstration, here's the Carl Zeiss Sonnar-T
>> 135mm f/1.8, which itself is no shoddy performer, on full 35mm format.
>>
>>
>> Pull the slider on the left from f/22 down to f/1.8 to see the effect of
>> diffraction on resolution. It improves down to about f/8, but doesn't
>> actually get any better going down to f/1.8 (certainly not in proportion
>> to the f/# as expected for just diffraction) and even degrades at lowest
>> f/#s. Even though some parts of the image, such as the centre, remain as
>> sharp at f/1.8 they aren't 4.5x sharper. So, for this 135mm f/1.8 the
>> effective diffraction limited f/# is around f/8.

>
>What does that say in the sense of resolving detail at f/1.8 on a FF
>24.6 Mpix frame?
>

That it isn't much better than resolving detail at f/8, and certainly
not 4.5x better.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Kennedy McEwen, Jun 30, 2011