32 MP 4/3" sensor - why not?

[Jay Turberville]

The yield from such a sensor would be very poor, so the cost per
sensor would be terrible. The defect rate would be at least four
times as high (assuming the same technology, each defective pixel has
four times the chance to affect a given sensor).

Typical sensors have a lot of defective pixels. The camera keeps a pixel map that works around the defects. I can imagine that defects other than pixels might hurt yields, but pixel defects shouldn't.

If this was a
feasible idea it would already have been implemented.

That's poor logic. Live view was easily feasible since the first Foveon based DSLR was made. But Sigma is still hasn't implemented that feature. What gets produced is determined by much more that simply what is feasible. The fact that it isn't being done does not mean that it isn't feasible.

The processing requirements (memory, speed, power consumption) would
also go up by a similar ratio. Want to choose between a DSLR with a
100 shot per battery yield or a DSLR with an enormous battery?

Even if the image processing did scale proportionally, much of a camera's power drain isn't proportional. Running the LCD and AF are huge power drains. Also, if a binning mode is included, then the camera could run at a lower resolution much the same as a lower resolution camera. Binning gives you a flexible option to trade speed for resolution. That's be nice.

It is currently impractical,

If by that, you mean it is uneconomical or that camera makers see more profitable alternatives, then I agree.

although once the resolution
out-resolves the lenses it will be interesting to see whether the
manufacturer's finally drop the MP race or try to produce better
lenses to make the resolution usable.

It is hard to say. We can speculate on the image quality from such a sensor. But my suspicion is that camera makers are actually testing the benefits of that kind of approach or already have. Keep in mind also, that the notion of an image sensor that would "oversample" the lens output is somewhat new. Photographers seem to be somewhat slow to respond to new concepts. So such a camera would really need the right marketing campaign to go along with it.

--
Jay Turberville
http://www.jayandwanda.com

 

[Ross]

Good question. Only thing I can think of is that I expect that 32MP
may outresolve the existing lenses anyway so going to 44mp may not be
a further advantage. But if that is not so, then I don't see any
reason not to.

In some ways if this path is taken you end up with effectively the
cameras taking on the characteristics of film - slower, hi resolution
or faster low resolution - pick your body rather than your roll.

The max line pair resolution would be around 218 lp/mm, so yes, the sensor would out resolve the best lenses, even at 50% MTF. The image data file would choke most current computers and the current printers couldn't demonstrate the detail gain in prints, except in the largest prints. That said, industrial and medical applications might benefit.
--
Bob Ross
http://www.pbase.com/rossrtx

 

[OM4ever]

I think for 4/3 in particular, there is a marketing angle to producing such
a niche item. It could prove the viability of the better lenses with future
higher MP bodies.

Jan

If by that, you mean it is uneconomical or that camera makers see
more profitable alternatives, then I agree.

Jay Turberville
http://www.jayandwanda.com

 

[Kyle Jones]

Such a sensor will be diffraction limited between f/2.8 and f/4, and any photo with fine detail would be diffracted to mush just when most Four Thirds lenses start hitting their peak resolution, at f/5.6 to f/8. So you'd want a new line of lenses with peak performance from f/2 to f/2.8 to get full resolution out of the sensor. Yeah. Good luck with that.
--
Street: http://www.wonderworks.com/streetphotographydigest.html
Flickr: http://www.flickr.com/photos/kyle_jones/

 

[DriftForge]

Actually diffraction effects would be physically no different than on a current 4/3 sensor - the sensor is the same size.

Also, the comments about the size of the sensor leading to higher defect rates - the sensor is no bigger than the current 4/3 sensor, so the defect rates would be the same. It may have to be manufactured on a finer process to keep circuitry sizes smaller, which may increase cost somewhat, but sensors have to be some of the simpler chips out there in terms of repetitive patterns. One step above memory chips. A side benefit would be lower power usage due to the finer process.

And unless there is a specific need for 5fps I don't see that you couldn't just slow down the capacity rate if you wanted to save on processing electronics. I'm not convinced that this is as big a cost as made out though.

 

[BJL]

Sure, the sensor would likely outresolve most if not all of the lenses in the system.

If a 32MP sensor distinctly outresolves all lenses for the 4/3 system, there would be no point to it, except marketing hype. And lenses for larger formats that need to cover larger image circles do tend to have lower resolution in lp/mm, and so a larger minimum useful pixel spacing. So the fact that lenses for far smaller digicam formats can about keep up with 2 micron pixel pitch does not show that the same is possible with the far larger 4/3" format image circle.

But if enough of the better 4/3 lenses can keep up with 32MP, I do see at least some market for such a high resolution, low ISO options, for the great amount of photography done at low ISO speeds. After all, high res., low ISO films like Velvia and Kodachrome have or had a loyal following. For those who also want high ISO sometimes, at worst carrying two bodies would be an option. (And down-sampling or binning could also offer an option of lower res. with lower noise at higher ISO speeds.)

 

[Kyle Jones]

... larger relative to them. So to get the most out of those pixels you've got to open up the lens more to reduce the CoC which necessitates those improbably sharp f/2 lenses I mentioned. Without them, the results won't look worse than what we have now, but it won't produce 32 MP of resolution either. If you don't get the resolution, then what's the point?
--
Street: http://www.wonderworks.com/streetphotographydigest.html
Flickr: http://www.flickr.com/photos/kyle_jones/

 

[Plusiotis]

future 4/3 sensor, or more realistically a 4/3 mount dslr, which will match/surpass this pixel count. What people forget is that the diameter of the 4/3 mount (and, indeed, of the rear element of all 4/3 lenses) is no smaller than that of the old OM (i.e. 35mm film sized) lens mount. At present there is no need to (but the capability to) increase sensor size, even if that means (like Canon and Nikon) having lenses and dslr bodies that are not fully interchangeable with one another.

 

[DriftForge]

Of course Olympus can produce a sensor to any size in the future - but that is not the point of this discussion. What we are observing about is that a 4/3" sensor could currently be produced at this resolution (variously 24-40MP) for a non prohibitive cost with useful performance that provides scope for photography using the 4/3 system that is not currently available. And more so that there would be sufficient interest in such a creation to make it commercially viable.

 

[esco]

pentax 645, 67II
--
Oldschool Evolt shooter

 

[Paul Shields]

... larger relative to them. So to get the most out of those pixels
you've got to open up the lens more to reduce the CoC which
necessitates those improbably sharp f/2 lenses I mentioned. Without
them, the results won't look worse than what we have now, but it
won't produce 32 MP of resolution either. If you don't get the
resolution, then what's the point?

Maybe the answer to get to 24 or 34mp is to move on from the Bayer filter. It's not the most efficient way of using up pixels. A Foveon filter at the same pixel density may overcome some of these issues?

 

[Jay Turberville]

Such a sensor will be diffraction limited between f/2.8 and f/4, and
any photo with fine detail would be diffracted to mush just when most
Four Thirds lenses start hitting their peak resolution, at f/5.6 to
f/8.

Nope. It just wouldn't work that way. Here is a sequence of image taken with a C7070 (7Mp 1/1.8" sensor) that has a slightly smaller pixel pitch than a 32Mp 4/3" sensor. The sequence steps through f/4, f/5.6, f/8 and f/11. The image at f/8 is quite usable - especially if you consider the samples shown are 100% pixels and that you'd have 32 million such pixels to build your picture on. And if the sensor didn't have an AA filter, it'd probably be sharper.

It is true that the image does start to noticably soften at f/8 and is quite soft at f/11. But I'd bet a 32Mp sensor on a 4/3" sensor will shot noticably more detail at f/11 than we get with an 8Mp 4/3" sensor. It certainly wouldn't be less.



Here's a Coolpix 8400 at f/5.1 (8Mp 2/3" sensor) vs. an E-330 with a 50mm ZD macro at f/8. F/5.1 isn't producing mush. F/8 would certainly be softer, but it wouldn't be mush either. Expecially if you have 32 million pixels instead of only 8 million.

So you'd want a new line of lenses with peak performance from
f/2 to f/2.8 to get full resolution out of the sensor. Yeah. Good
luck with that.

The idea wouldn't be to get the "full resolution" of the sensor. The idea would be to eliminate the AA filter (which reduces contrast in fine detail) and have a sensor that can record the "full resolution" of the lens.

--
Jay Turberville
http://www.jayandwanda.com

 

[Jay Turberville]

Sure, the sensor would likely outresolve most if not all of the lenses in the system.

If a 32MP sensor distinctly outresolves all lenses for the 4/3
system, there would be no point to it, except marketing hype.

I disagree. The point would be to oversample the lens image instead of having to use an AA filter. That way you get all the detail the lens can deliver. So what if it can't deliver all the resolution theoretically possible with a 32Mp sensor. It will deliver more detail.

And
lenses for larger formats that need to cover larger image circles do
tend to have lower resolution in lp/mm, and so a larger minimum
useful pixel spacing. So the fact that lenses for far smaller digicam
formats can about keep up with 2 micron pixel pitch does not show
that the same is possible with the far larger 4/3" format image
circle.

No, the fact that Olympus is still using a fairly agressive AA filter on its 10Mp sensors tells you that 10Mp isn't the limit. And a 32Mp sensor only has twice the linear resolution of an 8Mp sensor.

--
Jay Turberville
http://www.jayandwanda.com

 

[PDXguy]

Not bad...



--
Regards,
Steve

 

[DriftForge]

30 odd posts in, good discussion going on a reasonable possibility that hasn't been discussed recently and here we are, inanity raises its head.

You want a reasonable impression of an ISO100 image, I can post one from my 8080 at the same pitch. Or you can look a few posts up at the 8400 which from the description is similar. Now given the quality of 4/3's lenses and four years development in sensors, I couldn't match what the discussed sensor would produce.

 

[PDXguy]

Grow a sense of humor; it's Saturday night.

Besides, with a little Noise Ninja, that should clean right up.
--
Regards,
Steve

 

[Dylan Betts]

Sure, the sensor would likely outresolve most if not all of the lenses in the system.

If a 32MP sensor distinctly outresolves all lenses for the 4/3
system, there would be no point to it, except marketing hype.

I disagree. The point would be to oversample the lens image instead
of having to use an AA filter. That way you get all the detail the
lens can deliver. So what if it can't deliver all the resolution
theoretically possible with a 32Mp sensor. It will deliver more
detail.

And
lenses for larger formats that need to cover larger image circles do
tend to have lower resolution in lp/mm, and so a larger minimum
useful pixel spacing. So the fact that lenses for far smaller digicam
formats can about keep up with 2 micron pixel pitch does not show
that the same is possible with the far larger 4/3" format image
circle.

No, the fact that Olympus is still using a fairly agressive AA filter
on its 10Mp sensors tells you that 10Mp isn't the limit. And a 32Mp
sensor only has twice the linear resolution of an 8Mp sensor.

How do you know that Olympus is using a "fairly aggressive AA filter?"

 

[Jay Turberville]

How do you know that Olympus is using a "fairly aggressive AA filter?"

Can't say for sure that this is the case, I assume it mostly from the "per pixel" sharpness differences I've seen from other DSLR with similar resolution and from the general impression that most people seem to have.

--
Jay Turberville
http://www.jayandwanda.com

 

[Jonathan Demarais]

30 odd posts in, good discussion going on a reasonable possibility
that hasn't been discussed recently and here we are, inanity raises
its head.

I hope you heard the whoosh as the meaning behind his post went over your head.

--



'I cried because I had no E-3. Then I met a man with no E-510'

 

[Jogger]

Even if they could keep the noise in check, there would be zero dynamic range.. might as well use a webcam.