Does a 39.3mp Sensor Exist? [CR1]

[rs]

There are meant to be various resolutions out there, all moving us somewhere closer to an upper limit which isn't worth going beyond for a 35mm sensor. For most purposes, a 20 odd MP sensor is already good enough, but for those that want maximum resolution, where will it end up?

Even lenses which are razor sharp with current sensors between f2.8 and f11 are likely to start showing their flaws when the MP goes up. The current 20 odd MP sensors start to suffer from DLA (diffraction limited aperture, or softness due to the small aperture) at around f11 with any lens. If the sensor can has a higher MP count, that smallest aperture with full res drops down - the D800 has a limit of approx. f7.8 - and most lenses are at their sharpest at around f8 - so it takes a really good lens to get the best out of a sensor that dense. Canon are releasing some very good lenses just recently which appear to out resolve the current Canon sensors at f2.8, and will possibly out resolve 36MP at f2.8 - but you've still got a reduced useful operational window there.

Its a exponential curve, so according to my calculations a sensor which is past its optimal range at f5.6 has 70MP - f4 is 138MP, and f2.8 is 282MP. Even at 70MP, f5.6, its unlikely any lens will give 100% pixel sharpness at f2.8, and to even appear perfect at f5.6 will take some great glass. So to get the best out of that sensor you're probably limited to one or two lenses, at one aperture setting. Select any other aperture, and even if everything else is perfect (tripod mounted photo of a stationary subject), you're not getting full resolution. And the narrow DoF at f5.6 at those magnifications (why else do you want such a high res, other than to print massive and study it up close) largely renders the extra resolution useless. This is possibly taking it too far for any practical uses.

Any sensor intended to work well for video will ideally be a nice round upscale of the intended HD resolution. At 1x 1080p, filled out to a 3:2 frame, its 2.5MP. 2x, its 9.8MP. 3x, 22.1MP. 4x, 39.3MP. 5x at 61.4MP will not allow for a useful upscale of 4k, so such a high res is really only suitable for 1080p, and will likely be no better than 39.3MP. And 6x is 88.5MP - while it will downscale for video OK, it's too much for stills at its native resolution by my reckoning.

So is 39.3MP the ultimate resolution for a 36x24mm sensor designed with HD video in mind?

And is it as far as we need to go for stills, or will demand make some manufacturer take it higher still? Diffraction sets in at f7.5 at 39.3MP - not inconceivable for landscape and studio work, but would some users want more, considering the limitations it brings? If you need much more resolution to print at massive sizes and still look good when studying it with a magnifying glass, you're hitting the limits of the size of the sensor.

[dilbert]

That is the main reason the 5D MkIII is killing the D800 in the wedding market (well amongst the local populace of pros in my neck of the woods), a 5D MkIII RAW file is a couple of mb bigger than a 5D MkII RAW file, a D800 RAW is three times the size of a D700 one.

Strange for you to say that because wedding photographers generally have no time for dealing with RAW and thus they shoot JPEG.

I think you'll find that most if not all of the 5D3 wedding shooters already had Canon equipment.

[dilbert]

As some here probably already realize, 39.3 megapixels is kind of a magic number, since its resolution is probably 7680 (i.e.,1920 X 4 and 3840 X 2) horizontal by 5160 vertical. If you want a C300-style 2 x 2 binning sensor, except capable of being cropped to 3840 X 2160, that is the resolution you would want.

Good observation. A 4x4 HD 3:2 sensor has 39.3 MP; a 5x5 has 61.4 MP

I don't think that 5x5 would work. The bayer sensor uses a 2x2 block of two green, one blue and one red pixels. Using odd numbers would require taking pixels from an adjacent block. I don't think the 5D3 uses 3x3 binning, although the horizontal resolution is 5760 or 3 x 1920.

Many people (including myself) thought that the 5D3 would use 3x3 binning for video.

It doesn't.

[rs]

Yes, diffraction is happening in the lens regardless of the sensor, but the advantage of the higher MP sensor is lost much easier to diffraction. If you have two cameras, one 20MP and the other 282MP, and shoot both with a lens set with its aperture smaller than f11, there will be no difference in resolution between the two. The higher MP body won't be worse - it just won't have any advantage. But shoot at f5.6 with a great lens, and the 282MP sensor will be able to resolve 70MP of detail while the 20MP sensor is limited to 20. So while 282MP clearly holds an advantage with a great lens in that particular scenario, why bother with 282MP? Isn't the cut off of where a FF sensor should ever go to somewhere around 70MP or below? And if its a video optimised sensor, 39.3MP seems like the ultimate destination for FF if video remains based on a multiple of 1080p (4k or 8k), even hundreds of years from now.

File sizes are a major concern when using these large sensors. As you said, the resolution of the D800 is too much, too soon for event shooters with current computers.

I was looking more into how far it is worth going based on the laws of physics, assuming sensor tech and storage/processing all continue improving, making these resolutions easy to capture and post process.

[rs]

Yes, diffraction is happening in the lens regardless of the sensor, but the advantage of the higher MP sensor is lost much easier to diffraction. If you have two cameras, one 20MP and the other 282MP, and shoot both with a lens set with its aperture smaller than f11, there will be no difference in resolution between the two. The higher MP body won't be worse - it just won't have any advantage. But shoot at f5.6 with a great lens, and the 282MP sensor will be able to resolve 70MP of detail while the 20MP sensor is limited to 20. So while 282MP clearly holds an advantage with a great lens in that particular scenario, why bother with 282MP? Isn't the cut off of where a FF sensor should ever go to somewhere around 70MP or below? And if its a video optimised sensor, 39.3MP seems like the ultimate destination for FF if video remains based on a multiple of 1080p (4k or 8k), even hundreds of years from now.

File sizes are a major concern when using these large sensors. As you said, the resolution of the D800 is too much, too soon for event shooters with current computers.

I was looking more into how far it is worth going based on the laws of physics, assuming sensor tech and storage/processing all continue improving, making these resolutions easy to capture and post process.

There are very sound arguments for up to 400mp, but with big caveats. http://theonlinephotographer.typepad.com/the_online_photographer/2009/02/why-80-megapixels-just-wont-be-enough.html

Going to the extremes of a 400MP FF sensor will allow you to see 400MP of beautiful blur up close when viewed at 100%, unless a lens exists to resolve 400MP of resolution at f2.3 and wider apertures. Having said that, software enhancements such as SmartDeblur, if further refined could turn those extra blurred pixels into meaningful detail:

http://www.petapixel.com/2012/10/21/smartdeblur-does-science-fiction-esque-enhancing-on-blurry-photos/

I still reckon 20 odd MP is enough for most people, and 70MP without any artificial enhancement should be more than enough for any real life situation.

[art_d]

Going to the extremes of a 400MP FF sensor will allow you to see 400MP of beautiful blur up close when viewed at 100%, unless a lens exists to resolve 400MP of resolution at f2.3 and wider apertures.

I recall seeing on another forum that someone had tested the resolving power of a Canon 24TSEII on a Pentax Q with the rationale being that 12.4mp on the tiny sensor would correspond to a 360MP full frame sensor. The uptake of this experiment was that the 24TSEII obviously delivers the necessary resolving power. So I don't think it's that big of a stretch.

Granted, 300-400MP may not be practical for many photographers, but it certainly is possible with lenses that exist today. And we are likely to see many more lenses with improved resolving power in the future.

[Stu_bert]

Yes, diffraction is happening in the lens regardless of the sensor, but the advantage of the higher MP sensor is lost much easier to diffraction. If you have two cameras, one 20MP and the other 282MP, and shoot both with a lens set with its aperture smaller than f11, there will be no difference in resolution between the two. The higher MP body won't be worse - it just won't have any advantage. But shoot at f5.6 with a great lens, and the 282MP sensor will be able to resolve 70MP of detail while the 20MP sensor is limited to 20. So while 282MP clearly holds an advantage with a great lens in that particular scenario, why bother with 282MP? Isn't the cut off of where a FF sensor should ever go to somewhere around 70MP or below? And if its a video optimised sensor, 39.3MP seems like the ultimate destination for FF if video remains based on a multiple of 1080p (4k or 8k), even hundreds of years from now.

File sizes are a major concern when using these large sensors. As you said, the resolution of the D800 is too much, too soon for event shooters with current computers.

I was looking more into how far it is worth going based on the laws of physics, assuming sensor tech and storage/processing all continue improving, making these resolutions easy to capture and post process.

There are very sound arguments for up to 400mp, but with big caveats. http://theonlinephotographer.typepad.com/the_online_photographer/2009/02/why-80-megapixels-just-wont-be-enough.html

Going to the extremes of a 400MP FF sensor will allow you to see 400MP of beautiful blur up close when viewed at 100%, unless a lens exists to resolve 400MP of resolution at f2.3 and wider apertures. Having said that, software enhancements such as SmartDeblur, if further refined could turn those extra blurred pixels into meaningful detail:

http://www.petapixel.com/2012/10/21/smartdeblur-does-science-fiction-esque-enhancing-on-blurry-photos/

I still reckon 20 odd MP is enough for most people, and 70MP without any artificial enhancement should be more than enough for any real life situation.

The article linked by privatebydesign indicates that raw engines overcome some of the diffraction limitations, otherwise surely a 7D sensor would have the same problem as the D800 and be noticeably impacted by diffraction after f/8?

[rs]

Going to the extremes of a 400MP FF sensor will allow you to see 400MP of beautiful blur up close when viewed at 100%, unless a lens exists to resolve 400MP of resolution at f2.3 and wider apertures.

I recall seeing on another forum that someone had tested the resolving power of a Canon 24TSEII on a Pentax Q with the rationale being that 12.4mp on the tiny sensor would correspond to a 360MP full frame sensor. The uptake of this experiment was that the 24TSEII obviously delivers the necessary resolving power. So I don't think it's that big of a stretch.

Granted, 300-400MP may not be practical for many photographers, but it certainly is possible with lenses that exist today. And we are likely to see many more lenses with improved resolving power in the future.

If that's the case, my calculations are out. I was basing it on DLA as quoted on the-digital-picture, and found a website to allow you to show what various cameras pixels sizes are, compared to the Airy Diameter of a perfect lens at a given aperture: http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm

Anyway, to calculate the pixel size of a sensor in µm, it turns out the calculation is =36/(sqrt(MP/2*3), and the ratio between DLA and pixel size seems to be 1.6 (to match up with Bryan's calculations at TDP). So my calculation was DLA = 57.6/(sqrt(MP/2*3). Which is according to your above post about the Pentax Q1 and 24 TS-E II, is incorrect.

If that is the case, 39.3MP is just a stepping stone towards much higher resolutions for those that demand it. The next logical resolution for video will be 39.3MP x 4 to allow for perfect downsampling of 1080p, 2k and 4k - and that's 157.3MP if the sensor is scaled up to 3:2. And stills? Who knows? I had assumed we were closing in on the practical limits as governed by diffraction with this next generation of sensors, with a possible sensible upper limit of around 70MP.

However, the Nikon 70-200/2.8 with a Nikon V1 seeing the centre crop at an equivalent of just over 70MP does fit in with my calculations - the lens at f5 is stopped down sufficiently to get close to its sweet spot in terms of sharpness, but crucially not beyond f5.6 to get diffraction setting it. However, that a FF lens such as the Nikon 70-200 is good enough to resolve that much detail does surprise me. It would be interesting to know from anyone using a FF lens on a V1 if the results suffer when moving too far away from f5.6?

It would be interesting to see an f3.5 lens resolving, even in the centre of the frame, enough to look truly sharp at 360MP when viewed at 100%. According to my calculations (which may be out), 360MP would need f2.5 to avoid diffraction, and a perfect f3.5 lens will top out at 180MP of resolution.

Thanks!

[Daniel Flather]

I like that interesting lens on the EOS 3D......

That 40L lens image has been used on many rumour images, so don't hold your breath.

[LetTheRightLensIn]

Many people (including myself) thought that the 5D3 would use 3x3 binning for video.

It doesn't.

Then how come it doesn't have the nasty moire of the 5D2 and 6D?

It is a bit soft, but I'm guessing that is because the AA filter was designed for 1x1 single photosite sizes for stills and doesn't help any for 3x3 blocks so they do a little AA in software to mush away some of the aliasing from that??? Not really sure though.

[rs]

Many people (including myself) thought that the 5D3 would use 3x3 binning for video.

It doesn't.

Then how come it doesn't have the nasty moire of the 5D2 and 6D?

It is a bit soft, but I'm guessing that is because the AA filter was designed for 1x1 single photosite sizes for stills and doesn't help any for 3x3 blocks so they do a little AA in software to mush away some of the aliasing from that??? Not really sure though.

As a bayer sensor has 2x2 blocks to give the full colour info - one red, one blue and the two green, a 5D mk III 3x3 block downsampled into one video pixel, while better than what could be downsampled from a 21 or 20MP sensor (hence their use of pixel binning), it's not ideal. A multiple of 2x2 is theoretically much better, such as the 39.3MP 4x4 downsample.

I would have guessed some sort of of merging of the resulting pixels occurs in software, probably to do with compensating for the inconsistent colour info across each 3x3 block.