Sensor and 12 bits

[yb98]

I'm wondering why all sensors are 12 bits ?

It would be more logical to have 2 familiy of sensors :

8 bits : for people who shoot in jpeg

16 bits : for people who shoot in raw and like to get the best from their camera

Yacine.

[CyberDyneSystems]

It isn't really a matter of choice in that sens,. they are 12 bits because of the math involved in the way that the colors are filtered for the CCD/CMOS... errr.. :?

I don't know the math, nor can I find the site that explained it all... :roll:

Anyone?

[Jesper]

My "educated guess".....: :wink:

I think the sensors that are used in today's digital cameras simply are not sensitive enough to record more than 12 bits per pixel of information.

If a camera would convert the voltage measured at each pixel site with more than 12 bits per pixel, the extra bits would probably only contain noise, and not real image information. So that's my idea of why we currently don't have 16 bit per pixel sensors. In the future we will probably have better sensors with less noise and a better dynamic range, and then it will make sense to have more than 12 bit per pixel.

Why manufacturers don't make two types of sensors? I don't know... When silicon chips are manufactured, there are always some that are just a bit better than others. For example, with microprocessors, after they produce a number of processors, they are tested, and some pass the test for running on a higher speed, and some don't - so the best ones are sold, for example, as 3 GHz processors, while the less good ones are sold as 2 GHz processors (I don't know if it still happens like that, but it used to be this way in the time of the 386 and 486). I can imagine with sensors for digital cameras there would be something similar. A manufacturer could use the sensors that have more than a certain amount of noise for cheaper cameras, that only have 8-bit JPEG mode, and use the better sensors for more expensive cameras that have 12-bit RAW mode.

[robertwgross]

I think that sensor resolution of 12 bits is more what is easily achievable using existing semiconductor technology, rather than what the camera user marketplace demands. In math terms, 8 bits, 12 bits, and 16 bits are nice round numbers.

Following the actual sensor elements, the A-D converters were once able to do only 8 bits of resolution. Now they are easily able to do 12 bits. Probably if we go away and come back in a few years, they may be able to do more, like 16 bits.

However, personal computer technology (including graphics cards and displays) may have to improve to stay ahead of the camera technology. It wouldn't do much good to have a 24-bit-per-color image in your camera and not have anyplace to manipulate it conveniently.

---Bob Gross---

[theoldmoose]

Computers are already behind in this respect.

Current display adapters do 8-bit per pixel, at most (not counting any alpha hardware that may be there, but that doesn't really figure into this).

So, a so-called 32-bit display mode is really only 8-bits per R,G,B channel, and maybe 8-bits total for an alpha channel.

Whether or not it's useful to have more than that in a display adapter will depend heavily on monitor technology. At the moment, most LCD displays will only show about a 400:1 contrast ratio, which is only a bit better than 8-bits, anyway. CRT's are usually a bit better, but not by much. Even a 1000:1 ratio is not as good as 10-bits (1024 levels).

And high contast papers and printing really hasn't come close to 'real world' dynamic range, which can be 10-stops or better.

Aside from 12-bit cameras, some film scanners boast of 12- or 14-bit conversions (and seem to be able to deliver it).

Manipulating 12-bit data in 16-bit space, though, allows you some overhead, so you won't tend to clip or limit values while you do your mathematical manipulations.

In the end, though, you will have to flatten to 8-bits, just to output it to most devices.

[Guillermo Freige]

Also, sensors work in 12-bit using a linear space (gamma 1.0) and monitors and printers in sRGB or AdobrRGB colorspace use a non-linear space (usually gamma 2.2) so you actually need a 16-bit nonlinear space to map a 12-bit linear one without loosing intermediate values in highlights. That's why some people recommend to expose as near as highlight clipping as posible, because as the RAW colorspace is linear, half of the values are used by the last stop. Of course when you use JPEG, a lot of values are discarded in the middle to bright areas of the image, but not so many or none at all in the shadows, because very few values are present there and they can "fit" in 8-bit space.