You're still thinking in terms of having an image that's sharp down to the individual photosite. With resolution overkill, that's no longer a design goal, and crosstalk is actually not that bad a thing to have. It functions as a form of antialiasing.
Paul & Gordon,
Then doesn't that get back to my original statement. If some future type of imager is using such densly packed photosites that they no longer even attempt optical isolation then while it may have 50 million photosites it won't be close to that resolution. On top of that the lack of electrical isolation raises the noise floor further dimishing any gains made through a higher photosite count.
Also... we never addressed the Canon statement that they had trouble creating top grade lenses that could accurately resolve 16MP much less 50!
Back up a minute. You're still ignoring my core point: that we don't WANT a sensor that isolates every photosite. Because doing so brings with it all the issues I mentioned previously: aliasing, moire, interpolation guesswork, and so on.
Ask yourself one question: if you have a lens that can resolve to a 20 megapixel level, then which sensor will produce a better picture when printed or displayed at the same size: a 16 megapixel sensor or a 64 megapixel sensor? The answer is the latter.
There's a tradeoff to be made between sharpness and noise, and if we use lots of megapixels then we can make the tradeoff in an optimal way, but if we use barely sufficient megapixels, then the tradeoff gets made for us, with no choice. And then it starts with the moire and the aliasing and the oy vey.
Using too many megapixels is the only way to get an image that's as good as your lens. People keep thinking that the goal is to be as good as your sensor. I say that's a fundamentally dumb and backwards approach. The picture comes from the lens, and all the sensor can do is mess it up. Let's get the sensor out of the way, so sensor resolution issues just aren't a factor. Then the lens determines the picture.
Early compact disk players had issues with high frequencies because they had to sharply filter them to keep the sampling noise out of the output. This produced distortions in the treble. Later CD players used "oversampling" to overcome this problem. By interpolating mathematically to double or quadruple the number of digital samples, they removed the need for harsh analog filtering.
But once that was done, did they stick with quadruple oversampling because it was perfectly adequate? No, they went for 64x and 128x oversampling. Just because they could.
Same here. There's no reason not to sample the light at every two microns instead of every twelve, even though there's no great gain in picture quality as a result. At the moment there are cost reasons why they don't do that, but in time those costs will disappear. The price of a sensor is going to be determined mainly by its overall size, not its density.
Eventually, a 100 megapixel camera is practically guaranteed to happen... even though the final picture quality may be only mildly better than what we have now. There's simply no reason
not to do it.
