Kodak boosts digital camera sensitivity

Sounds interesting.

http://news.com.com/Kodak+boosts+digital+camera+sensitivity/2100-1041_3-6190890.html?tag=nefd.lede




Another edit...

Nice idea, but it reduces the real resolution of the sensor further...

Doesn't that effectively happen with noise refduction algorithms anyway?

yes but the point being that you don't apply NR at low ISOs, so you get your full resolution capability back. This pattern will be awesome for security cameras and the like which must work well in low light. Not so sure it will work well for artistic photography in normal light though.

Of course back in the "good old days" you didn't have to worry that a 1600 ISO film had poor resolution in good light, because in good light you loaded a 100 ISO film with uber fine grains.

Aha. My apologies. I thought the same chip could be reconfigured to high and low sensitivity pattern in camera. Now that would be useful.

unless I misunderstand how these filters work, the colour filters are physical gels which cover a pixel, and they are basically a part of the sensor. You'd basically need interchangeable sensors to be able to swap them.

They are fixed to the surface of the CCD, CMOS or LMOS. Exchangeable sensors would be rather expensive, but it would be possible. Imagine having it as an exchangeable back, but then you'd be in a whole new realm where you could choose a nice Pentax body and load it up with a FujiFilm CCD

yes it will reduce the colour resolution of the camera further, but will actually slightly improve the broadband luminance resolution

its actually more similar to the human eye (there are way more rods than cones in the eye)

it will improve B&W shots a lot too.

it seems like an excellent compromise, although a prism and 3CCD system would be better (but presumably too bulky and expensive for a DSLR)

I've always thought that the bayer array was a poor solution because it wastes 2/3 of the light...

I can imagine some CA problems due to the arrangement of photo detectors of different colours. From looking at the diagram it takes 16 photo detectors to generate one pixel, where as currently you only need 4. This means you'd need a 40MP sensor to get the same resolution as a current 10MP sensor. Surely the extra density would completely undo the noise reduction?

yes it also looked to me as if a better arrangement of the coloured pixels could be achieved.

I think de-mosaicing is more sophisticated than that, each pixel presumably looks at all of its neighbours to work out what colour it should be.

however in a standard bayer array every pixel is next to a pixel of each other colour, seems like they should go for that in the high sensitivity array too, maybe by reducing the number of green pixels? why not 50% white and split the RGB equally? I'm sure they have their reasons, although they may not reflect the priorities we might have.

There are various de-mosaicing algorithms which look at extended areas beyond the neighbouring pixels. This still produces some errors and if you are processing 16 points to produce one you will get quite a few aliasing problems especially with specific colours. I can imaging some banding at 100% and a few other problems. You really need more pixels for the new array to work well.

On a Bayer every vertex meeting point of four photoreceptors is an imaginary pixel which needs information from a RGB, adding a fourth colour or just a 'white' to a standard array would be a good step as well without diminishing resolution and it can be used for noise reduction purposes, just not as effectively.

I think changing the shape of photoreceptors to better stacking hexagons and working on new pattern and algorithms would be better. You get less aliasing (if implemented correctly), more input for each pixel calculation (because you have more vertices) and therefore more 'imaginary' pixels. This part of the way FujiFilm deal with the problem and look at shots from an S5Pro at ISO1600 or even 3200, quite clearly superior.