645D lost the competition

Same pixel count and same image sensor size as the newly announced Mamiya MF digital. But Mamiya has 16 bit color and penatx only has 14 bit (here are the Pentax 645D specification: http://www.infoborder.com/Digital_Cameras/Pentax/645d.php ). How did pentax lose 2 bits of color if the image sensor is the same which is seems likely it is? Maybe this is how pentax is able to sell the camera for so cheap compared to Mamiya.

I can't see that 16 vs 14 bit makes much difference.
They use a different sensor in the Pentax to the Mamiya.

http://www.pentax.jp/english/news/2010/201008.html for all the official details of the 645D.

and both will have to be converted down to 8 bit or less for display or printing.

I'll reserve my judgement until I see some photos from a production camera. The competition you mention hasn't really started yet.

not to mention that Mamiya is over TWICE the price of the Pentax, for $20 grand I should hope it makes the tea and composes the images itself.

The K10D use a whopping 22-bit A/D converter for example so it must be good

And the 16 vs 14 bit argument is about the A/D converter (as I understand) - so since 14 bits is already more than you get out of the sensor, 16 bit is basically just digitising the noise in a bit more detail?
Neal

the sensor is analogue, it produces a voltage signal which is measured by the A/D converter. The sensor itself doesn't have a "native" bit level.


The other question which as far as I know is yet to be answered is how many bits do you actually get in the RAW files. As Stefan reminded us above, the K10D has a 22bit A/D converter but only outputs 12bit RAW files which limits the benefit.

Sorry, I wasn't clear. As I understand it, attempting to digitise the sensor output with 'too many' bits doesn't get you any more 'real' image data, just better digitised noise which isn't particularly helpful. I don't exactly know what 'too many' is in this context, but I believe it's < 14, so 14 vs 16 (vs 22) is moot except for marketing.
Neal

sortof,

the way it works is:

light falls on the sensor and interacts with the semiconductor, creating electron-hole pairs. Readout is in the form of an analogue voltage proportional to the number of electrons generated.

According to the spec sheet for the Kodak sensor which matches the specs of the 645D most closely (we don't know if that's the actual sensor to be used, but the specs will be similar) it has a saturation charge of ~41 thousand electrons.

Since you can't have part of an electron, there are 41000 possible levels of output from that sensor. 2^14 is ~16000 levels while 2^16 is ~65000 levels. So assuming that sensor (which may or may not be the one in use in either camera), 16bit should be better than 14 although not quite as much as might first appear.

Interestingly, though I don't know the specs for the K10D's sensor, that sensor has a similar pixel size and therefore probably similar saturation charge, meaning K10D's 22bit A/D converter is almost certainly mostly hype as there are only about 15.5bits of data there to digitise.


So in short, yes, you're right, we're too close to quantum physics to pretend that true analogue signals exist in nature at this point, but the step up from 14 to 16 bit does appear to be justifiable from the sensor specifications.


What I do find interesting is this really highlights for me how wrong people are when they say things like "yes but CCDs will always keep getting better". Actually it looks like we're getting dangerously close to the theoretical limits of what can be done with some of this stuff.

The new sensors are already on the way, it does not work like a bucket like the CMOS or CCD... well it does but the bucket is very small and it simply counts the number of times a bucket will be filled, you can have huge dynamic range this way.

are you talking about photon counting imaging sensors? that would be very interesting, do you have a link for that?

Now that you mention it I've seen something like that for X-rays. At the moment the spatial resolution is diabolical, no where near good enough for visible light imaging, but of course CCDs started out that way so perhaps in a few years....

here's what I've seen

http://www.dectris.com/sites/pilatus1m.html

obviously this one is designed for X-ray wavelengths. The pixels are too large and the price too high (I enquired about buying one for my lab, ~£300,000 for a 1MPix one :shock but photon counting detectors do work nicely and remove the problem of saturation.

Of course you still hit the same noise floor when there aren't enough photons as you do with a CCD.

Here you go link
At the moment they use 256 grey scales and when the bucket is charged up they cut it and then you count how many buckets are full.
So you need 256 pixels for one pixel on the screen so it's not so interesting at the moment.

If you read the comments they are already speculating that you could count the number of times the bucket is filled, now if they could do that, that would be highly interesting.

Oh your machine looks much more interesting and here I come with a simple SD card thing that has something like a gigapixel for a few bucks

thanks for the link, that's a facinating idea, will be interesting to see how it develops!