Canon EOS 5Ds and 5DsR - 50.3MP with and without AA filters

[Bart_van_der_Wolf]

Can anyone explain what is meant by "a much stronger set of colour filters"?

Hi Gary,

Many camera/sensor makers strike a compromise between sensitivity and color accuracy. The sensitivity is boosted by using less selective, more transparent, RGB filters in the Bayer CFA. Those less selective filters will have a larger overlap between the color transmission bands, which makes it more difficult to extract the correct colors. So 'stronger' should be read as 'more selective'.

Whether the filters are indeed more selective, is of course still subject to speculation, as is the supposed announcement of a new camera.

Cheers,
Bart

[ErikKaffehr]

Hi,

I guess it is a bit more complex than that. Colour filters are a compromise.

Best regards
Erik

Hi Gary,

Many camera/sensor makers strike a compromise between sensitivity and color accuracy.

Cheers,
Bart

[Paul2660]

I read the following on the Northlight blog about the 5Ds,

"there is a suggestion that the relatively low max ISO is from a much stronger set of colour filters than usual - more welcome news if so"

Can anyone explain what is meant by "a much stronger set of colour filters"?

Thanks

The stronger set of color filters, points to a Foveon type chip.  This was also mentioned quite a bit around Photokina times.  Canon does have a patent for this type of chip.  So far the Foveon style chip has not shown good high iso results, and this may point to the high iso of 6400. 

However since a lot of the talk comes from Canon Rumers, many have forgotten this post from about a week ago.

http://www.canonrumors.com/2015/01/a-sony-canon-sensor-partnership-mentioned-again-cr1/

With either solution the new cameras will be I am sure highly regarded by most photographers. 

Next week will very interesting, and don't forget about Sony's new A7rII or A9, whatever it will be called.

Paul

[Gary Ferguson]

Many camera/sensor makers strike a compromise between sensitivity and color accuracy. The sensitivity is boosted by using less selective, more transparent, RGB filters in the Bayer CFA. Those less selective filters will have a larger overlap between the color transmission bands, which makes it more difficult to extract the correct colors.

Thank you Bart!

[LKaven]

The stronger set of color filters, points to a Foveon type chip.

I'd be /very/ surprised if this happened. 

[Paul2660]

I'd be /very/ surprised if this happened. 

I agree, but it would be a much bigger step, at least to me.

Paul

[Bart_van_der_Wolf]

I agree, but it would be a much bigger step, at least to me.

As unlikely as that direction is, it would certainly make the file size explode ...

No, at approx. 8688 x 5792 pixels, with a sensel pitch of approx. 4.14 micron, (false color) aliasing artifacts are already less likely to occur. Nevertheless, I'd still prefer a version with AA-filter (OLPF), the EOS 5Ds. Digital Signal Processing (DSP) dictates the necessity for a Low-pass filter to reduce (color) aliasing artifacts when using a discrete sampling device.

Cheers,
Bart

[dreed]

With either solution the new cameras will be I am sure highly regarded by most photographers. 

Lets not count our chickens before they've hatched.

Thus far, nobody has seen what comes out of the camera.

It will be interesting to see how it measures up at DxO.

[eronald]

Lets not count our chickens before they've hatched.

stop crying fowl-play!

Edmund

[Gel]

The twist will be when it turns out the R has a Foveon sensor and the 50mp is the total of three layers. 

That would be the best Camera trolling ever.

[Paul2660]

The twist will be when it turns out the R has a Foveon sensor and the 50mp is the total of three layers. 

That would be the best Camera trolling ever.

That would still be a totally awesome sensor at full frame 35mm. Be it 20 or 25MP.

Paul

[Bernard ODonovan]

Is the video 4K?

I did not think it would, if true, here is the latest rumor specs from the original source in Japan:

"- Sensor Effective Pixels 50.6MP. The total number of pixels 53MP CMOS
- 5DS R is a low-pass filter is disabled
- RAW (50MP), M-RAW (28MP), S-RAW (12.4MP)
- Media CF (UDMA7), SD / SDHC / SDXC (UHS-I)
- Dual DIGIC6
- Crop 1.3x and 1.6x
- Finder penta prism, 100% field of view, magnification 0.71 times, the eye point 21mm
- Electronic Level
- Grid display
- AF 61 points (41 points cross type). EV-2 support.
- ITR AF
- Anti-flicker
- Time-lapse movie
- Bulb timer
- Live View, the contrast AF (face recognition)
- 150,000 pixel RGB-TR metering sensor. 252 zone TTL metering
- EOS iSA system
- ISO100-6400 (extended with ISO50 and ISO12800)
- The shutter speed is 30 seconds -1/8000 seconds. Synchro is 1/200 sec
- Continuous shooting 5 frames / sec.
- Video 1920x1080 30fps (ALL-I or IPB)
- LCD monitor 3.2 inches 1.04 million dot
- Mini HDMI output terminal. External microphone terminal
- Battery LP-E6N / LP-E6
- The size 152 x 116.4 x 76.4mm
- Weighs 930g (CIPA guidelines). 845g (body only)"

[Guillermo Luijk]

No, at approx. 8688 x 5792 pixels, with a sensel pitch of approx. 4.14 micron, (false color) aliasing artifacts are already less likely to occur. Nevertheless, I'd still prefer a version with AA-filter (OLPF), the EOS 5Ds. Digital Signal Processing (DSP) dictates the necessity for a Low-pass filter to reduce (color) aliasing artifacts when using a discrete sampling device.

The idea that the lack of AA filter is better than having it has already extended in the photographic community. Discussing about the reasons why the AA should be there with non experts in signal theory is exhausting. I think this has become an opportunity for camera makers to charge you more for the non-AA version, when it actually should be the opposite.

Anyway, 50Mpx require such a weak AA filter that the differences in practice should start to become almost negligible.

Regards

[dwswager]

The idea that the lack of AA filter is better than having it has already extended in the photographic community. Discussing about the reasons why the AA should be there with non experts in signal theory is exhausting. I think this has become an opportunity for camera makers to charge you more for the non-AA version, when it actually should be the opposite.

Anyway, 50Mpx require such a weak AA filter that the differences in practice should start to become almost negligible.

I am certainly not a Signal Theory expert.  All I know is that neither my D7100, nor my D810 have AA filters and they are the 2 sharpest DSLRs I've owned or played with.  And neither exhibit moire or other artificats.  Whatever Nikon is doing, I recommend, they keep doing it!  The D800e was odd in that it actually had an AA filter, but then the effects were filtered back out.

[eronald]

The idea that the lack of AA filter is better than having it has already extended in the photographic community. Discussing about the reasons why the AA should be there with non experts in signal theory is exhausting. I think this has become an opportunity for camera makers to charge you more for the non-AA version, when it actually should be the opposite.

Anyway, 50Mpx require such a weak AA filter that the differences in practice should start to become almost negligible.

Regards

I once had an interesting exchange with the Fuji guys who claimed that they were immune to aliasing by virtue of their sensor design. I asked what happens when you have a repetitive signal eg. at twice the Nyquist limit. And they replied that is impossible, the lens will filter that out

The AA filter has been such a mainstay of dSLR design that I am not sure we understand the *practical* implications of removing it. One should remember that much of signal theory was developed by the telephone guys, and they were interested in reducing the bandwidth before encoding for transmission... an issue that is not really present in high-end photography although doubtless relevant in video.

Edmund

Full disclosure - I'm not an expert, I just own an original edition of Shannon & Weaver that had been thrown away by an antiquarian bookstore

[Bart_van_der_Wolf]

The AA filter has been such a mainstay of dSLR design that I am not sure we understand the *practical* implications of removing it.

Hi Edmund,

It's quite simple. No AA filter (Optical Low-pass Filter, or OLPF in our case) means more aliasing artifacts. Since the Bayer CFA samples Red and Blue more sparsely than Green, the aliasing will create False color artifacts when demosaicing the different sized aliasing patterns.

Besides that, the filter stack is part of of the modern optical design, so leaving out parts will create chromatic and spherical aberrations, and that's why the AA-less versions of the cameras still need some blank material of identical refractive index (hopefully Anti-Reflection coated or kitted together) to maintain a proper optical path.

Cheers,
Bart

[Guillermo Luijk]

I am certainly not a Signal Theory expert.  All I know is that neither my D7100, nor my D810 have AA filters and they are the 2 sharpest DSLRs I've owned or played with.  And neither exhibit moire or other artificats.  Whatever Nikon is doing, I recommend, they keep doing it!  The D800e was odd in that it actually had an AA filter, but then the effects were filtered back out.

If I were to choose between an AA or non-AA version, I'd choose the non-AA version (well, only if it were at the same price or cheaper), but this is just because I'm just a hobbyist and I like to play with 100% crops, and mainly because I don't need photography to pay the bills. If I were a professional, I'd certainly not take risks and I'd choose the AA versión. Even if moiré or artifacts are not very likely to appear, a client requiring or even appreciating the extra sharpness on the non-AA sensor is much, much less likely to show up.

BTW a part of the perceived extra sharpness is just non-filtered high frequency interference, not useful signal. A similar thing happens when noise is added to an image: the second image looks sharper, I just added gaussian monochrome noise to the former:




Bart and others have shown a good amount of siemens stars where it was clear how non-AA sensors create false high frequency artifacts when the scene contains fine detail.

Regards

[eronald]

Bart and others have shown a good amount of siemens stars where it was clear how non-AA sensors create false high frequency artifacts when the scene contains fine detail.

Regards

I don't believe in "magical" high-cut filters; there are always side-effects, including an attenuation below Fc, phase rotations etc.

Especially when as Bart helpfully points out we have two different sampling frequencies at work. In fact, someone should convince me that the alisasing will not persist because of these factors for any AA filter that does not cut off *well* below the frequency defined by the inverse of the BB or RR *diagonal* distance. (there may be a factor 2 somewhere there). The Bayer asymmetry is horrid, and the filter slope will possibly affect differently oriented color structures differently, creating problems.

I do understand some of the maths involved, and I'm not so sure that there is any perfect solution within the context of Bayer.

Maybe someone can run the appropriate simulators, and check, but using Bayer, and the kernels of real-world AA filters, and real sensors.

Edmund

[Josh-H]

The twist will be when it turns out the R has a Foveon sensor and the 50mp is the total of three layers. 

That would be the best Camera trolling ever.

Just keep in mind that Canon hasn't actually said anything yet - this is just 'leaked' rumour. 'If" the camera did turn out to be 3 layer Foveon (which it won't BTW  ) it wouldn't be Canon that spread the initial misleading info...

[Phil Indeblanc]

I heard it would be $8K! I hope not.
If it is a out-house sensor, wouldn't the price also be closer to the numbers we are used to..sure a bit higher as it is 50 vs 36mp, but I don't think they should pass roughly $4K price point(?)
The D810 is under $3k now. How much a difference is the extra mp going to make in price? Given the pixel quality is equal.