[bossa]

I do wonder how much better this can get. Remember the backlit has less influence when pixels grow larger. No idea why anyone wants 63 megapixels, unless your images are up for large streetadvertising. I mean all of them.

My computer has a hard time rendering my D800 images - it was struggling when I tried to process a 645Z RAW file - so I'm not in the market, just interested.

[eurostar]

I do wonder how much better this can get. Remember the backlit has less influence when pixels grow larger. No idea why anyone wants 63 megapixels, unless your images are up for large streetadvertising. I mean all of them.

A 63 megapixel sensor, by 4x4 binning, would allow for a 15 MP sensor without interpolation of colors. I would like it.

[RonHendriks1966]

My computer has a hard time rendering my D800 images - it was struggling when I tried to process a 645Z RAW file - so I'm not in the market, just interested.

This is why I lost interest in the K-3. On my computer fine, but on my laptop I was waiting. Sold it this week, so now I'm fine with K-5 and K-01.

[gazonk]

A 63 megapixel sensor, by 4x4 binning, would allow for a 15 MP sensor without interpolation of colors. I would like it.

I guess you meant 2x2 binning?

Your point is valid, though. I also wonder if it would make sense to make bayer filters with more than three colors in order to better capture "difficult" colors like spectral violet? For instance 9 different colors in 3x3 "hyper pixels".... you'd need a 90 mp camera to get those pixel binned and get a 10 mp image.

[RonHendriks1966]

A 63 megapixel sensor, by 4x4 binning, would allow for a 15 MP sensor without interpolation of colors. I would like it.

Well you could ask them to make a 16 megapixel sensor at Full Frame that gives that same IQ.

[Winder]

I'm rather interested to see sensor tests of the Samsung NX1 as I'm hoping that the 28MP backside lit sensor tech offers a quantum leap in the noise dept. Just imagine a FF sensor made out of that... ..63MP! The lenses look pretty nice too.

A lot of people are definitely interested in the technology. The the photodiode layer moved in front of the metal wiring they can put larger phtotdiodes on the same size sensor. The 28MP Samsung should have close to the same "pixel" size as the 16MP Sony CMOS. It should outperform the Sony FF 24MP in terms of both color and noise, but I don't expect a quantum leap.

If you are looking for a quantum type leap then the organic sensors being developed in partnership between Fuji and Panasonic are more interesting. The APS-C organic sensors they have developed have a 4 stop advantage over the current 36MP FF Sony. Fuji has been delaying the X-Pro 2 release for this technology.

[falconeye]

A 63 megapixel sensor, by 4x4 binning, would allow for a 15 MP sensor without interpolation of colors. I would like it.

The colors must always be interpolated. If you take the R, G, and B subimages from a raw and overlay them to create an image of clear colors and half resolution, it looks awful. Tried it.

I also wonder if it would make sense to make bayer filters with more than three colors in order to better capture "difficult" colors like spectral violet?

Doesn't make sense.
The best possible color sensor works with 3 color filters each replicating one of the three human eye's color sensitivities as close as possible. Might work for bee eyes though

The 28MP Samsung should have close to the same "pixel" size as the 16MP Sony CMOS.

That difference is egalized by microlens technology. The net difference will be marginal.

However, BSI diodes have a better angular response curve, even with microlenses. This could help solve the current problems with the corners of wide angle lenses on high resolution mirrorless cameras such like the A7r.

The APS-C organic sensors they have developed have a 4 stop advantage

There is no such thing as a 4 stop advantage.

Quantum efficiency (QE) of current sensors already is at 50%, only 1 stop left. Another stop may come by replacing color filters by another technology to register photon wavelength. And that's it.

There are false claims from companies trying to attract venture and stock capital. They are typically based on creation of many electrons per photon. But this doesn't help if you already measure every electron created. QE is the conversion rate of photons to electrons.

Fuji uses this definition:

The sensitivity of image sensors refers to the conversion ratio of light into electric signals

But without increasing the percentage of photons converted at all, it won't change the image noise. They're cheating and for a reason ...

---

Last edited by falconeye; 09-26-2014 at 09:07 AM.

[Clavius]

I'm obviously no engineer. But would it be possible to have "software-SR" in an FF-format-camera operating in APS-C mode by making use of the extra space?

[Kunzite]

Not unless you can make the images from separate "sub-exposures" (i.e. stack several shorter exposures) at a very high rate. Otherwise, with the current technology there's no way to differentiate when a photon hit a certain "pixel".
I'm ignoring the mechanism used in some compacts (see Pentax Pixel Tracking), based on raising ISO, as it doesn't get good results.

[Clavius]

Not unless you can make the images from separate "sub-exposures" (i.e. stack several shorter exposures) at a very high rate. Otherwise, with the current technology there's no way to differentiate when a photon hit a certain "pixel".
I'm ignoring the mechanism used in some compacts (see Pentax Pixel Tracking), based on raising ISO, as it doesn't get good results.

I was assuming that it doesn't need to know when a photon hits a certain pixel, but just needs to compensate for movement. Just like SR does.

[Kunzite]

The problem is, you cannot emulate what the SR (in-body or in-lens) does via software. With an optical or mechanical stabilization system, the subject's image position is (more or less) fixed in relation to the sensor. A sensor cannot "track" the subject's image moving over in one "large" exposure, so the idea was to split the exposure into several shorter sub-exposures (for which the camera movement won't be an issue), and align the resulting images on the flight.
But I think this is impractical, and I don't expect a real life implementation. For software-based solution, we're stuck with trying to reconstruct a movement-blurred image.

I'm curious if anyone has a better idea/opinion, the subject is definitely interesting.

[falconeye]

The problem is, you cannot emulate what the SR (in-body or in-lens) does via software.
...
I'm curious if anyone has a better idea/opinion, the subject is definitely interesting.

That's correct. However ...

SR via software is nothing else than bursting away at higher iso and faster shutter speed, then later align & stack in SW like PhotoAcute. That can always be done by using burst mode.

And you are correct, in oder to simulate optical SR, one needs higher readout rates than available today.

However ...

If I read the recent interview about the Samsung NX1 correctly, then Samsung managed to do trick around that limitation. They seem to have embedded an on-chip processor which is able to do some limited hardwired processing w/o the need for readout. They use it for their Autoshoot feature analysing the image data at 240fps, e.g., triggering when a bat hits the baseball, automagically. For the time being, that still can't emulate optical SR because that would require additional storage for the image data. However, add 50-100 MB of SDRAM to a sensor chip and it would become feasible. However, I'd expect optical SR for Pentax to be the simpler route by FAR.

[eurostar]

I guess you meant 2x2 binning?

Your point is valid, though. I also wonder if it would make sense to make bayer filters with more than three colors in order to better capture "difficult" colors like spectral violet? For instance 9 different colors in 3x3 "hyper pixels".... you'd need a 90 mp camera to get those pixel binned and get a 10 mp image.

Yes, it was 2x2.

[Kunzite]

@Falk:
Yes, this is exactly what I'm talking about, including how I envision it working (i.e. on-chip support with the need to transfer to and store in an intermediate memory). The performance and storage requirements are not small, Autoshoot probably doesn't transfer/process the entire image (subsampling it instead) but the "software SR" will have to, at similar or even faster frame rates.

It cannot work (as a SR replacement) by using burst mode, because the frame rate is simply too small; but of course, the principle is the same.

[Digitalis]

If you take the R, G, and B subimages from a raw and overlay them to create an image of clear colors and half resolution

With RAW files the colour channels are incomplete. As an experiment a few years ago I took three images of a still life setting using a Leica monochrom using R,G,B filters - it was a real pain to find filters with the right spectral bandpass . I combined the images in Photoshop and produced an image that gave the image qualities of foveon sensors - with Leica glass.

Quantum efficiency (QE) of current sensors already is at 50%, only 1 stop left. Another stop may come by replacing color filters by another technology to register photon wavelength. And that's it.

Or you could get rid of the filters entirely.