24mp sensor for m43? 3.2micron pixel pitch!

Started May 9, 2013 | Discussions
RicksAstro
RicksAstro Veteran Member • Posts: 3,805
Re: Let's hope not........

oklaphotog wrote:

seachicken2000 wrote:

oklaphotog wrote:

But if those were iso 3200 files, the 24mp Nex-7 file would show more noise as we're slightly pushing both files past their original size.

I find I get similar quality high ISO images from my 16 and 24M pixel NEXs (using Lightroom).

For example below (the best I could do from the DPR RAWs for each image) at ISO3200 I'd say the NEX-7 image has about the same level of noise, but it's finer grained, and the image is a touch more detailed.

But this doesn't really show it, you've picked a bright area of the scene instead of the shadows where the differences are much greater on any camera. Also you have used noise reduction. Most people going for ultimate image quality will use 0 luminance and try to stay away from more than 25 color and trade a little grain for better resolution.

For many of us we run from NR, as quick as possible. Here's a more realistic scenario between the two in the shadows. Left image is from the nex-6 uprez'd to the same size as the nex-7 which is on the right. ISO 3200 and no noise reduction. This shows what is really going on in the sensor.

ISO 3200 Nex-6 on left scaled up to the same size as Nex-7 on the right.

Not a huge difference, but it is there. Probably a stop or so.

Since the NEX-7 is indeed slightly worse above ISO3200 on DXOMark, it stands to reason it's slightly worse in the underexposed areas at ISO3200.

But you really can't use DPR's examples to judge this.   They do not control the light reaching the sensor, so comparisons are really not valid.   In your example above, the shutter for the NEX7 is 1/1000 and the NEX6 is 1/800.  And we really have no idea of the studio light intensity. So who knows how much light was actually hitting the sensor (which is the main determining factor for noise in these mostly ISO-less cameras)

Also, I've found upresing to the larger pixel count really isn't a great method to visually judge noise.   Of course the NEX7 shows more high frequency noise since it's at full res...the nex6 couldn't possibly show that high frequency noise.    I find it a more effective tool to downres either to a common resolution or to the resolution of the lowest.

And even with all of that, there is at most 1/3-1/2 stop difference.   Just look at the same area with the NEX6 set to ISO6400...it has more noise even at the 100% pixel level.   Viewed at the same magnification, the nex-7 is clearly better, so the difference is far less than one stop.

Finally, I've found noise reduction software if far more effective with higher res files when viewed at the same magnification as lower res files, keeping more of the fine detail.

Bottom line is I think DXOMark's graphs reflect reality, and the performance of these cameras is similar through most of the range, with a slight (about 1/3 stop) advantage to the NEX6 above ISO 6400.

-- hide signature --
seachicken2000
seachicken2000 Veteran Member • Posts: 3,234
Re: Let's hope not........

oklaphotog wrote:

But this doesn't really show it, you've picked a bright area of the scene instead of the shadows where the differences are much greater on any camera. Also you have used noise reduction. Most people going for ultimate image quality will use 0 luminance and try to stay away from more than 25 color and trade a little grain for better resolution.

Even with the noise reduction I have applied to the NEX-7 image there's more detail remaining than in the NEX-5n image with no noise reduction. You can back off the NR to trade grain for further detail in the NEX-7 image if you wish.

It's the subject you want sharp and detailed in an image, and usually it's going to be lit, so the performance on a non-shadow area of the image at high ISO is relevant.

For many of us we run from NR, as quick as possible. Here's a more realistic scenario between the two in the shadows. Left image is from the nex-6 uprez'd to the same size as the nex-7 which is on the right. ISO 3200 and no noise reduction. This shows what is really going on in the sensor.

I don't think this is realistic, unless you like chroma noise in the shadows. Kill the chroma noise with 25 reduction, and leave the luminance noise at 0 as you suggest, and you get parity in terms of noise (NEX-7 left, NEX-5n right):

-- hide signature --

"The only thing that gets in the way of a really good photograph, is the camera"
Norman Parkinson

 seachicken2000's gear list:seachicken2000's gear list
Sony Cyber-shot DSC-R1 Fujifilm FinePix S2 Pro Sony Alpha NEX-5 Sony Alpha NEX-7 Sony Alpha a7R +21 more
Franka T.L. Veteran Member • Posts: 8,148
The constraint and the display

is not on the sensor.

Like it or not , there is several thing to consider, a sensor inside a DC of any type is build in ( unless you shoot Medium or Large format Digital Back ) and that's t. It much be made to work on all possible shooting environment placed ( high ISO , high humidity , low light, bright light etc ... ) and increasing the MP obviously is placing too much constraint on this performance curve

The other thing to consider, the display of photo and yes the display of Video captured. In near term future and mid term 4K is likely the case, so let's take the high road and say 4096 for 4K ( Many 4K is just quad HD which is 2X2 x [1920X1080] ) well for such a capture in 16:9 we need only 9.44 MP which by simple math one would deduce even the 16MP now we have is more than enough. But that's only part of the story; well; until the coming of truly usable full spectrum sensor , those capture is still de-mosaic(ed) from the Bayer pattern capture. So for no nonsense approach to getting more , the 2 way to do it is either:  use no AA filter ( can do ) or one can binn ( minimal 2X2 grid of photosite per resultant pixel on image ) to get full spectrum recording . This is more so for Video than for Photo. And of course all this mean 4X more photosite , so even for a 16:9 HD we are talking about a  37.8 MP , and that infer a sensor with 45.3 MP , way more than the demanded 24MP. Ok enough tech mumble jumble ... the point is .. 24MP just would not give much advance over the 16MP ( on a per axial basis its only barely making perceivable advance ), and question is - can you get that difference on your home HD monitor or that computer monitor of yours, certainly I do not think so for your tablet ( even if you are using the retina display iPad )

Read the rumor about the 50MP FF sensor from Canon and of that from Sony then and now, well there is a reason why and most of that had to be with the Cine / Video instead of Photography and not about capturing more with those nice fat L or Zeiss lens.

-- hide signature --

- Franka -

RicksAstro
RicksAstro Veteran Member • Posts: 3,805
Nokia 808 41 mpix smartphone
4

Taken to the extreme for a case study is the Nokia 808 Pureview 41 mpix phone camera.   It had 1.4um pixels, which would be the equivalent of having a 115 mpix m4/3 camera or a 440 mpix full frame!   Surely if there were an inherent IQ disadvantage to those tiny pixels, it would surface on this camera.   But it blows the rest of it's competition out of the water (DPR's words)

As processors become faster, the native resolution will become just another spec.   The user will be able to choose whatever output resolution (RAW equivalent or JPG) they want..

Take a read of DPRs review here:

http://www.dpreview.com/articles/8083837371/review-nokia-808-pureview

Here's a great excerpt from the Q&A section:

"How will you send the message that the 41MP sensor is for better quality images, not bigger images?

I think the easiest way to get the message of the image quality is just to see the images!

In the default mode the camera captures 5MP images, and the quality of those 5MP images is just amazing. We need to separate two things: output resolution - what is right number of megapixels for the outputting image files - and input resolution - what is the resolution of the sensor that is used to capture the image.

When we have the huge input resolution, we have a lot of advantages: oversampling improves the signal to noise ratio, dynamic range, and the output files have very natural non-processed look without signs of interpolation, sharpening or noise filtering. In addition, we can zoom without reducing the lens' maximum aperture.

Starting with a larger number of smaller photosites but outputting at a lower resolution allows us to increase the signal to noise ratio, too, too. By combining the signals from 7 neighboring diodes, you can get the same signal to noise ratio as you would from a photodiode seven times larger (if you do it right). Of course, we could have just used fewer, larger photosites, but then you would lose all the advantages of oversampling, plus the ability to zoom."

-- hide signature --
tedolf
tedolf Forum Pro • Posts: 26,562
Secret truth........
2

oklaphotog wrote:

A2T2 wrote:

oklaphotog wrote:

RicksAstro wrote:

tedolf wrote:

noise and diffraction effects will suffer.

And for what gain?

We already had all the resolution we needed when we hit 8 mp.

Tedolph

When viewed at the same size, diffraction actually has LESS of an effect with more pixels than with less, so you do still gain a little even in diffraction limited land (although not as much as the difference in the number of pixels).

And as I've mentioned, when viewed at the same size, noise is not worse either with the latest of today's technology.

So you really lose nothing (except file size) and gain the potential to crop or enlarge.

Sure noise can be worse. Take the NEX-7 at 24mp and the NEX-5/6 at 16mp. At high ISO's the NEX-7 has worse noise than the 16mp cams when viewed at the same size, especially printing 12x18@300dpi or bigger.

and yet the new nikon d5200 is the best aps-c yet?

I said can be. Not all sensors are equal. The D5200 sensor is newer. But the 24mp and 16mp Sony sensors are of the same generation more or less, using the same technology. Obviously as tech gets better well have a better noise threshold per the same pixel size. But we are pretty close to hitting a brick wall right now as far as noise levels, which probably won't be significantly improved with out engineers really thinking outside the box and coming up with something very new and different.

I would bet that a 34mp Toshiba sensor using the same tech as the 24mp Toshiba sensor in the D5200 would produce more noise when viewed or printed large. Now if you just downres it to weeny sized prints or monitor viewing, sure they will look more or less equal.

Here it is.......

But who wants a high pixel count camera for monitor viewing or small prints like 8x10's. The whole reason for a high pixel count is to print bigger without having to use as much interpolation.

So if you are not printing gigantic there is no reason to $$ for higher resolution sensor.

The reality is that for most folks 16mp is more than adequate for the sizes they print.

You mean 8 mp.

But I like big prints.

How big?

Bigger than 16" x 20"?

12mp is more than adequate for that.

In the case of the Sony comparison, when uprezed to a 20x30" print at 300dpi, there is no visual difference in resolution when viewing the print even up close.

hardly anybody prints that large.

I have only printed that large twice in my life, for a trade show.

But if those were iso 3200 files, the 24mp Nex-7 file would show more noise as we're slightly pushing both files past their original size.

Tedolph

 tedolf's gear list:tedolf's gear list
Olympus PEN E-P5 Olympus M.Zuiko Digital ED 9-18mm F4.0-5.6 Olympus M.Zuiko Digital ED 40-150mm 1:4-5.6 Olympus M.Zuiko Digital 14-42mm 1:3.5-5.6 II R Samyang 7.5mm F3.5 UMC Fisheye MFT +8 more
tedolf
tedolf Forum Pro • Posts: 26,562
NEX 6 looks.......
1

oklaphotog wrote:

seachicken2000 wrote:

oklaphotog wrote:

But if those were iso 3200 files, the 24mp Nex-7 file would show more noise as we're slightly pushing both files past their original size.

I find I get similar quality high ISO images from my 16 and 24M pixel NEXs (using Lightroom).

For example below (the best I could do from the DPR RAWs for each image) at ISO3200 I'd say the NEX-7 image has about the same level of noise, but it's finer grained, and the image is a touch more detailed.

But this doesn't really show it, you've picked a bright area of the scene instead of the shadows where the differences are much greater on any camera. Also you have used noise reduction. Most people going for ultimate image quality will use 0 luminance and try to stay away from more than 25 color and trade a little grain for better resolution.

For many of us we run from NR, as quick as possible. Here's a more realistic scenario between the two in the shadows. Left image is from the nex-6 uprez'd to the same size as the nex-7 which is on the right. ISO 3200 and no noise reduction. This shows what is really going on in the sensor.

ISO 3200 Nex-6 on left scaled up to the same size as Nex-7 on the right.

Not a huge difference, but it is there. Probably a stop or so.

better.

More like film.

NEX 7 looks digital.

Tedolph

 tedolf's gear list:tedolf's gear list
Olympus PEN E-P5 Olympus M.Zuiko Digital ED 9-18mm F4.0-5.6 Olympus M.Zuiko Digital ED 40-150mm 1:4-5.6 Olympus M.Zuiko Digital 14-42mm 1:3.5-5.6 II R Samyang 7.5mm F3.5 UMC Fisheye MFT +8 more
Detail Man
Detail Man Forum Pro • Posts: 16,801
Re: Nokia 808 41 mpix smartphone

RicksAstro wrote:

Taken to the extreme for a case study is the Nokia 808 Pureview 41 mpix phone camera.   It had 1.4um pixels, which would be the equivalent of having a 115 mpix m4/3 camera or a 440 mpix full frame!   Surely if there were an inherent IQ disadvantage to those tiny pixels, it would surface on this camera.   But it blows the rest of it's competition out of the water (DPR's words)

As processors become faster, the native resolution will become just another spec.   The user will be able to choose whatever output resolution (RAW equivalent or JPG) they want..

Take a read of DPRs review here:

http://www.dpreview.com/articles/8083837371/review-nokia-808-pureview

Here's a great excerpt from the Q&A section:

"How will you send the message that the 41MP sensor is for better quality images, not bigger images?

I think the easiest way to get the message of the image quality is just to see the images!

In the default mode the camera captures 5MP images, and the quality of those 5MP images is just amazing. We need to separate two things: output resolution - what is right number of megapixels for the outputting image files - and input resolution - what is the resolution of the sensor that is used to capture the image.

When we have the huge input resolution, we have a lot of advantages: oversampling improves the signal to noise ratio, dynamic range, ...

How does that work ?

and the output files have very natural non-processed look without signs of interpolation, sharpening or noise filtering. In addition, we can zoom without reducing the lens' maximum aperture.

Starting with a larger number of smaller photosites but outputting at a lower resolution allows us to increase the signal to noise ratio, too, too. By combining the signals from 7 neighboring diodes, you can get the same signal to noise ratio as you would from a photodiode seven times larger (if you do it right).

Not improved, then ?

Of course, we could have just used fewer, larger photosites, but then you would lose all the advantages of oversampling, plus the ability to zoom."

Canonero Regular Member • Posts: 423
Nahh, on APS-C camras, megapixel war is over

I don´t believe that Olympus or nikon or sony, Canon still have a bit of megapíxel to move from 18 to something 22-24Mp.

Althougth 24Mp from Oly sensor would be fantastic, the most of people CAN´T see or understand the fantastic word CROP. I have several ton of weight on Canon lens and bodys, to make potos on street or mountain my canon hear is out of point, too big, too weight and eye effect.

Now, i will buy the next month the E-P5 + V-F4 + several lens: fantastic hear.

With my Canon 400mm 5.6L lens atachment to the Oly, i get a 800mm f5.6L lens stabilized, fantastic hear for static birding, i will try this combo. I have the Sigmonster with eos 1 body, but for walk on the mountain is out of question.

RicksAstro
RicksAstro Veteran Member • Posts: 3,805
Re: Nokia 808 41 mpix smartphone

Detail Man wrote:

RicksAstro wrote:

Taken to the extreme for a case study is the Nokia 808 Pureview 41 mpix phone camera.   It had 1.4um pixels, which would be the equivalent of having a 115 mpix m4/3 camera or a 440 mpix full frame!   Surely if there were an inherent IQ disadvantage to those tiny pixels, it would surface on this camera.   But it blows the rest of it's competition out of the water (DPR's words)

As processors become faster, the native resolution will become just another spec.   The user will be able to choose whatever output resolution (RAW equivalent or JPG) they want..

Take a read of DPRs review here:

http://www.dpreview.com/articles/8083837371/review-nokia-808-pureview

Here's a great excerpt from the Q&A section:

"How will you send the message that the 41MP sensor is for better quality images, not bigger images?

I think the easiest way to get the message of the image quality is just to see the images!

In the default mode the camera captures 5MP images, and the quality of those 5MP images is just amazing. We need to separate two things: output resolution - what is right number of megapixels for the outputting image files - and input resolution - what is the resolution of the sensor that is used to capture the image.

When we have the huge input resolution, we have a lot of advantages: oversampling improves the signal to noise ratio, dynamic range, ...

How does that work ?

and the output files have very natural non-processed look without signs of interpolation, sharpening or noise filtering. In addition, we can zoom without reducing the lens' maximum aperture.

Starting with a larger number of smaller photosites but outputting at a lower resolution allows us to increase the signal to noise ratio, too, too. By combining the signals from 7 neighboring diodes, you can get the same signal to noise ratio as you would from a photodiode seven times larger (if you do it right).

Not improved, then ?

Of course, we could have just used fewer, larger photosites, but then you would lose all the advantages of oversampling, plus the ability to zoom."

I can't answer your comments since these were from a Nokia Rep.

But on your last question, I think the current state of the art shows that the number of pixels doesn't have a large correlation with S/N at the image level, so no, S/N isn't improved, but it's also not necessarily worse like has been implied numerous times on this thread.

So the user has the choice of the trade-off of resolution/enlarging/cropping/"digital zoom" vs. S/N at whatever resolution they wish to view.    With fewer pixels, the user has less flexibility.

Every shot doesn't need to take advantage of every pixel for it to be an advantage in flexibility.

-- hide signature --
Detail Man
Detail Man Forum Pro • Posts: 16,801
Re: Nokia's Claims

RicksAstro wrote:

Detail Man wrote:

RicksAstro wrote:

Taken to the extreme for a case study is the Nokia 808 Pureview 41 mpix phone camera.   It had 1.4um pixels, which would be the equivalent of having a 115 mpix m4/3 camera or a 440 mpix full frame!   Surely if there were an inherent IQ disadvantage to those tiny pixels, it would surface on this camera.   But it blows the rest of it's competition out of the water (DPR's words)

As processors become faster, the native resolution will become just another spec.   The user will be able to choose whatever output resolution (RAW equivalent or JPG) they want..

Take a read of DPRs review here:

http://www.dpreview.com/articles/8083837371/review-nokia-808-pureview

Here's a great excerpt from the Q&A section:

"How will you send the message that the 41MP sensor is for better quality images, not bigger images?

I think the easiest way to get the message of the image quality is just to see the images!

In the default mode the camera captures 5MP images, and the quality of those 5MP images is just amazing. We need to separate two things: output resolution - what is right number of megapixels for the outputting image files - and input resolution - what is the resolution of the sensor that is used to capture the image.

When we have the huge input resolution, we have a lot of advantages: oversampling improves the signal to noise ratio, dynamic range, ...

How does that work ?

and the output files have very natural non-processed look without signs of interpolation, sharpening or noise filtering. In addition, we can zoom without reducing the lens' maximum aperture.

Starting with a larger number of smaller photosites but outputting at a lower resolution allows us to increase the signal to noise ratio, too, too. By combining the signals from 7 neighboring diodes, you can get the same signal to noise ratio as you would from a photodiode seven times larger (if you do it right).

Not improved, then ?

Of course, we could have just used fewer, larger photosites, but then you would lose all the advantages of oversampling, plus the ability to zoom."

I can't answer your comments since these were from a Nokia Rep.

But on your last question, I think the current state of the art shows that the number of pixels doesn't have a large correlation with S/N at the image level, so no, S/N isn't improved, but it's also not necessarily worse like has been implied numerous times on this thread.

Yes, given a comparable Fill Factor, (image-level) Photon Shot Noise is comparable, independent of Photosite size - because it is a property of the incoming light itself.

What I don't get is how (image-level) Dynamic Range manages to improve (due to cancellation of its random components) unless Photosite Full Well Capacity could somehow either be held constant, or at least held to be larger in value relative to the reduction in photo-active area.

Eric Fossum wrote:

... it is easier to design higher full well into a bigger pixel than a smaller pixel. Full well in the photodiode is mostly area x per area capacitance x voltage. The latter two are the same so it is easier to do with larger pixel area. But, normally one designs the output node for a particular conversion gain (uV/e-) and the maximum voltage swing therefore sets the full well and nominally independent of pixel area. It is a complicated design tradeoff.

http://www.dpreview.com/forums/post/38721415

So the user has the choice of the trade-off of resolution/enlarging/cropping/"digital zoom" vs. S/N at whatever resolution they wish to view.  With fewer pixels, the user has less flexibility.

Yes, that is true in cases where the option of using all of the smaller Photosites remains viable where it comes to potentially lower Dynamic Range and what certainly are higher Diffraction consequences.

Every shot doesn't need to take advantage of every pixel for it to be an advantage in flexibility.

But, when sensor/post cropping, the increased spatial frequency resolution is diminished, the image-level Signal/Noise Ratio is reduced (due to the fact that a smaller sensor/image area is utilized), and any lower Dynamic Range resulting from smaller Photosites continues to degrade performance.

BTW, I calculate a Photosite size of around 3.0 Microns (as opposed to 3.2 Microns) for 24 Mp.

Detail Man
Detail Man Forum Pro • Posts: 16,801
Re: Scaling doesn't work

EEmu wrote:

It's almost the first rule of engineering: you cannot simply scale something.

So you scale up the RX100 sensor, approximately doubling its size to be the same as a m43 sensor.  Let's ignore all light gathering concerns and take a quick look at how this affects the sensor design:

a) Readout circuitry is per column so we have 40% more readout circuits, each of which must service 40% pixels (i.e. rows).  This means, in short, higher read noise or reduced performance. In practical terms, it'd likely be a bit of both.

b) Power consumption of a CMOS sensor works out to be roughly per pixel so it'll produce twice the heat while maintaining the same thermal sinking properties (being a m43 sized sensor in a m43 sized body). This means it will run hotter and create more noise.

c) Semiconductor cost is driven by both area and defect rate.  Defect rate is driven by area (bigger -> more likely to defect) and feature size.  Doubling the area of the RX100 (i.e. scaled to m43 size) we both halve the number per wafer and double the likelihood of a defect. The overall expectation is that the cost will be roughly quadrupled.

Of course, there are even more issues past that, but hopefully that gets the idea across.  In short: scaled up sensors don't mean equivalently scaled performance.

Your points made above seem valid to me. Periodic as well as random components of Read Noise would likely increase relative to Full Well Capacities. Increased heat dissapation would increase Shot Noise arising out of reverse-biased leakage currents ("thermal noise" sources). Manufacturing complexities and costs would increase, and manufacturing yields would indeed decrease.

A 24 Mpixel Micro Four Thirds image sensor would require around an 18.35% reduction in Photosite height/width relative to 16 Mpixel image-sensors found in the GH3 and EM5.

JamesMortimer Senior Member • Posts: 1,492
Re: 24mp sensor for m43? 3.2micron pixel pitch!

3.2 micron?

It will be about as good as a pin sensor bridge camera...

Sensor size is really only important for the number of pixels in can carry at a large enough sensor site size.

A FF 21 mp camera might have 45 micron site size; change that sensor to a aps-c and it drops down to perhaps 23 microns.
Make it a pin head size, still keeping the 21 mp and the site size will be about 1 micron.

Bigger sites = better low-light ability, lower noise, less texture smearing.. generally better IQ all over (depending on other factors).

OP A2T2 Regular Member • Posts: 278
Re: 24mp sensor for m43? 3.2micron pixel pitch!

JamesMortimer wrote:

3.2 micron?

It will be about as good as a pin sensor bridge camera...

Sensor size is really only important for the number of pixels in can carry at a large enough sensor site size.

A FF 21 mp camera might have 45 micron site size; change that sensor to a aps-c and it drops down to perhaps 23 microns.
Make it a pin head size, still keeping the 21 mp and the site size will be about 1 micron.

Bigger sites = better low-light ability, lower noise, less texture smearing.. generally better IQ all over (depending on other factors).

Nikon d800 are 4.88µm! d7100 3.9? omd 3.8.

Anders W Forum Pro • Posts: 21,468
Re: 24mp sensor for m43 ? 3.2micron pixel pitch !

RicksAstro wrote:

Detail Man wrote:

...

As the extinction spatial frequency of the lens-system reduces (as the Wavelength multiplied by F-Ratio product increases), the above described (Nyquist limit) advantage of smaller photosites due to higher spatial sampling frequency begins to disappear. As well, ratio of the area under the compared composite system MTF responses (below the Nyquist spatial frequency limit of the larger sized photosite sensor) also begins to diminish - to the point where the ratiometric differences approach what are practical insignificance. Thus, "more pixels equals more detail" is a bit of a "mantra" ...

Yes, there are some cases (usually at the extremes of aperture and/or lens quality) where more pixels will not give you any significant detail.   Perhaps this statement would be a little more informational...

"More pixels will never yield less detail and will usually yield more."

Yes. The whole truth and nothing but the truth in a single sentence. Mission accomplished.

 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +28 more
John Sheehy Forum Pro • Posts: 20,471
Re: 24mp sensor for m43? 3.2micron pixel pitch!

One_Oldman_4U wrote:

photofan1986 wrote:

Are you not satisfied with 16 megapixels?
I don't care about 24 megapixels. I would rather like to see an updated version of the 16 megapixels sensor, with even less noise, even better DR and a global shutter.

I agree! I rather have the same 16MP that less noise and better IQ or a 14MP with larger pixels for better low light photography.

The difference between the tiniest and largest pixels in use these days, as far as high-ISO noise per unit of sensor area is concerned, is a lot smaller than you would think, and the difference that does exist may exist partly because the tiniest pixels are generally only used in products with low budgets, using cheaper support components like power supplies.

John Sheehy Forum Pro • Posts: 20,471
Re:More pixels does not mean more noise
1

RicksAstro wrote:

I say bring on the resolution and ditch the AA filters!   Giving an option to do on-camera resolution reduction for raw would be handy as well for those that don't wish to have the larger size images.

I would not say ditch the AA filters.  You would need about 250MP to ditch them safely for 4/3, 1GP for FF.  I would say make them electronic (vary their distance from the sensor or emulate them by shaking the sensor at ultrasonic speeds) with user-selected implementation, based on f-stop, focus distance, and lens (and maybe shutter speed.

Cameras without AA filters are only without problems when used for things like macro with large DOF, or shooting super-tele hand-held, like using a relatively slow (f/5.6) 300 or 400mm lens hand-held with a 2x converter, in which case the resolution limits of the lens will do sufficient anti-aliasing.  I would not want to use a camera without an anti-aliasing filter to shoot architecture or city-scapes with a very sharp lens at f/2.8!  There's such a thing as sharp, and too sharp with gross spatial artifacts.

John Sheehy Forum Pro • Posts: 20,471
Re: 24mp sensor for m43? 3.2micron pixel pitch!

Anders W wrote:

If we stop down further, say to f/16, both cameras will lose resolution due to diffraction. Again, I don't know the exact numbers, but the resolution figures might now be something like 800 for the D800 versus 600 for the D700. Based on these numbers, the D800 would lose more than the D700: 1000 - 800 = 200 versus 700 - 600 = 100. And in this sense, it is in fact correct to say that diffraction will have a stronger impact on the camera with higher pixel count than on the one with lower. However, and that's the really important point, the D800 will still be ahead of the D700. The resolution advantage of the D800 will be smaller and smaller as you continue to stop down beyond the peak aperture, but it will still have the advantage.

Another advantage is that when you are in the range of diffraction where it causes a black/white transient to take several pixels to occur, the more sensor resolution you have, the more you KNOW that a pixel that stands out against its neighbors is bogus, and you can reject it completely, or trim it back to the highest or lowest value that it is optically possible for it to be.  You can not do this with confidence when the finest noise and the finest details are in the same spectrum!

Anders W Forum Pro • Posts: 21,468
Re:More pixels does not mean more noise

John Sheehy wrote:

RicksAstro wrote:

I say bring on the resolution and ditch the AA filters!   Giving an option to do on-camera resolution reduction for raw would be handy as well for those that don't wish to have the larger size images.

I would not say ditch the AA filters.  You would need about 250MP to ditch them safely for 4/3, 1GP for FF.  I would say make them electronic (vary their distance from the sensor or emulate them by shaking the sensor at ultrasonic speeds) with user-selected implementation, based on f-stop, focus distance, and lens (and maybe shutter speed.

Cameras without AA filters are only without problems when used for things like macro with large DOF, or shooting super-tele hand-held, like using a relatively slow (f/5.6) 300 or 400mm lens hand-held with a 2x converter, in which case the resolution limits of the lens will do sufficient anti-aliasing.  I would not want to use a camera without an anti-aliasing filter to shoot architecture or city-scapes with a very sharp lens at f/2.8!  There's such a thing as sharp, and too sharp with gross spatial artifacts.

What about moiré reduction via software? The E-M5 has a sufficiently weak AA filter that it is not unusual to see color moiré. But the LR tool for moiré reduction takes it away without much problem as far as I can tell. Can't say I have found a need to use it very frequently but it seems to work well when I need it.

 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +28 more
JamesMortimer Senior Member • Posts: 1,492
Re: 24mp sensor for m43? 3.2micron pixel pitch!

A2T2 wrote:

JamesMortimer wrote:

3.2 micron?

It will be about as good as a pin sensor bridge camera...

Sensor size is really only important for the number of pixels in can carry at a large enough sensor site size.

A FF 21 mp camera might have 45 micron site size; change that sensor to a aps-c and it drops down to perhaps 23 microns.
Make it a pin head size, still keeping the 21 mp and the site size will be about 1 micron.

Bigger sites = better low-light ability, lower noise, less texture smearing.. generally better IQ all over (depending on other factors).

Nikon d800 are 4.88µm! d7100 3.9? omd 3.8.

Nope:

From snapsort -

Nikon D800 - Pixel size        23.8 µm²
Canon 5D2 - Pixel size        41.1 µm²
Nikon D7100 - Pixel size        15.3 µm²

The more pixels you put into a fixed size, the smaller they have to be - there's a balance between pixel size and overall sensor size.
And note that the D7100 has worse low-light ability than the 5D2.

A FF sensor with 1Mp and 800 micron pixel size might be useless, but so would be a FF sensor with 200Mp and a site size of 1 micron.

RicksAstro
RicksAstro Veteran Member • Posts: 3,805
Re:More pixels does not mean more noise

John Sheehy wrote:

RicksAstro wrote:

I say bring on the resolution and ditch the AA filters!   Giving an option to do on-camera resolution reduction for raw would be handy as well for those that don't wish to have the larger size images.

I would not say ditch the AA filters.  You would need about 250MP to ditch them safely for 4/3, 1GP for FF.  I would say make them electronic (vary their distance from the sensor or emulate them by shaking the sensor at ultrasonic speeds) with user-selected implementation, based on f-stop, focus distance, and lens (and maybe shutter speed.

Cameras without AA filters are only without problems when used for things like macro with large DOF, or shooting super-tele hand-held, like using a relatively slow (f/5.6) 300 or 400mm lens hand-held with a 2x converter, in which case the resolution limits of the lens will do sufficient anti-aliasing.  I would not want to use a camera without an anti-aliasing filter to shoot architecture or city-scapes with a very sharp lens at f/2.8!  There's such a thing as sharp, and too sharp with gross spatial artifacts.

Giving the option would be best.  I really liked the D800E better than the D800...noticeable increase in per pixel sharpness, so less post-sharpening was required (thus less high frequency noise enhancement).  A 24 mpix m4/3 would be even less prone to artifacts.

-- hide signature --
Keyboard shortcuts:
FForum MMy threads