Diko said:
Don Haines said:
Diko said:
there is already lossless jpg ver9 something... could be implemented in any of the next canon bodies (DIGIC 7 or 8 perhaps)
There is also a rumour that Canon has lossless RAW file format.
The problem with any lossless compression is that you end up with large file sizes... If lossless jpg isn't much of a savings over RAW, why bother?
Good to know
I mentioned it only due to the chit chat about DR and JPG.
jrista said:
If by QIS you are referring to Eric Fossum's Digital Film Sensor (DFS), that is a very old concept. Almost a decade old now, given this original paper:
http://ericfossum.com/Publications/Papers/Gigapixel%20Digital%20Film%20Sensor%20Proposal.pdf
I read that many, many years ago. Very intriguing concept...however it doesn't mean that you actually have a gigapixel sensor. The notion of a digital film is that the sensor works more like actual film which has silver halide grains, wherein the "jots" combine to make up large digital sensor "grains". Under lower illumination where there are fewer incident photons, one jot strike within the region of a grain would "illuminate" the entire grain as if each jot had received a photon. Grains remain large, resolution remains low, SNR is high, noise is low. Technically speaking, this isn't all that different from downsampling a high ISO image in post.
Under high illumination, where photon strikes are frequent, most jots would receive photon strikes. By employing a mechanism to "divide" digital grains, one could dynamically increase resolution, since smaller grains with fewer jots could still achieve a higher SNR. It's most definitely an intriguing concept, but it is also requires technological capabilities beyond what we are currently capable of (at least, as far as image sensor fabs go). Jots are considerably smaller than your average pixel...they would have to be close to deep red wavelength (somewhere between 750 and 800nm...current APS-C pixels are 4000-5000nm, current full frame pixels are 6000-9000nm).
To make an ideal Digital Film Sensor, I'd combine the Jot concept with the Titanium Nitride superconducting material and microwave comb readout to produce a sensor with infinite dynamic range, exact color replication, and effectively the highest resolution possible for an image sensor. The TiN technology is still pretty new, and pixel size is much larger than a jot (the only existing sensor is 44x46 pixels in size), and it still requires cooling in a dewar jar. But it would probably be the best sensor on earth.
Yes I do mean the same concept. These days Fossum calls it QIS. As for the thin film added to APS CMOS innovation from Canada IMO you have the whole concept wrong... No matter how he calls his pixels he claims of gathering 90%of the incident light and what is more important his intend is NOT to put it through ADC descrete process ergo the gigabytes. But even I could be wrong since he and his fellows are researching it as we speak.
The Q.E. is indeed high. I don't know about 90%, even with a BSI design unless he is supercooling, there is going to be a certain amount of loss due to dark current.
Having a high Q.E., however, does not change the notion of digital grains. In the presence of low light, you have low incident photon counts. The whole entire DFS/QIS design is based not just on jots, but on the fact that jots are organized into dynamic grains. In low light, all it takes is ONE jot to receive a photon in a grain for the ENTIRE grain to be activated. Let's say grains start out containing 400 jots each (20x20, a 16µm pixels...HUGE). Lets say were shooting in very low light, starlight. The moment one jot in each 20x20 size grain receives a few photons (lets say 50% Q.E., so two photons), then all 400 of those jots are marked as active! So, under low light, it might seem as though you actually received 800 photons, rather than just two! Big difference...you are now simulating the reception of a lot of light, however it is at the cost of resolution. At 16µm a grain, your resolution is going to be pretty low by modern standards...roughly 3.375mp.
Now, lets say a crescent or half moon comes out, and we take the same picture again. We have about two to three more stops of light. Instead of two incident photons, we now have ~8 incident photons per grain. Lets say a dynamic grain division is set at 8 photons. Once our jots receive and convert eight photons, our grains all split. We now have four times the resolution (10x10 grains, or 100 jots per grain, four times as many grains). We have a stronger signal overall, but roughly the same signal per grain as we did before. However we now have an image with four times as many megapixels, 13.5mp to be exact.
Now a full moon is out, and we take the same picture. We have another two stops of light. We get about 32 incident photons. Our grain size is now 5x5, or 25 jots per grain. Our resolution has quadrupled again. Same overall SNR, but our image resolution is 54mp.
This is what Eric Fossum has designed. A totally dynamic sensor that adjusts itself based on the amount of incident light, maintaining relative signal strength and SNR regardless of how much light is actually present. It does this by dynamically reconfiguring the actual resolution of the device...very low light, very low resolution, low light, low resolution, adequate light, good resolution, tons of light, tons of resolution. Technologically it is pretty advanced, conceptually it is relatively strait forward.
I've greatly exagerrated the scenario above...you wouldn't be able to have 54mp under moonlight. You would probably have something closer to 0.8mp under starlight, maybe 3mp under full moonlight, 13.5mp under morning or evening light, and maybe finally be able to achieve 54mp under full midday sunlight.
Diko said:
jrista said:
Canon is not in the image sensor market. Canon is in the photography market.
....
Canon won't be doing any kind of medium format anything any time soon.
Now , about the last statement are you sure?
Absolutely. I'm 100% sure. It makes no sense for Canon to try to break into a niche market that already has not only it's dominant players, but dominant players with a HELL of a LOT of loyalty among their customers. There have been Canon MF rumors for years. I remember reading MF rumors here back in the 2005 era. Nothing has ever come of them, despite how often Northlight tends to drag the subject back out.
The only way Canon could make a compelling entry into MF is if they launched an entire MFD system. Cameras with interchangable backs, image sensors that at least rival but preferably surpass the IQ of the Sony MF 50mp, a wide range of extremely high quality glass (they are certainly capable here, but it still is a MASSIVE R&D effort), and a whole range of necessary and essential accessories like flash. Canon has to do this all UP FRONT, on their own dime, to cover the massive R&D effort to build an entirely new system of cameras that can compete in an already well established market.
Now, they've done that once. They did it with Cinema EOS. But the cinema market is a lot broader with more players, and is a significant growth market with the potential for significant long-term gains, even for a new entrant like Canon. The medium format market is not a growth market. It's a relatively steady market, that has its very few players and it's loyal customers. Since there are so few players who already dominate the market, breaking in for a new player like Canon would be a drain on resources, and there is absolutely zero guarantee of any long-term payoff.
So, yes, I'm sure. Canon won't be offering a medium format camera any time soon.
Just do a surgery to implant a card slot in the skull, and we have images with the same quality as the human eye. ;D And with 20 stops of dynamic range. 
