Why 3 CCDs?
 

Canon and Nikon are making single-chip DSLRs so good with precise colors that they replace the 35mm film cameras already. Why the TV world still need 3 CCDs to get good colors?

( One can't even make prints out of TV cameras! )

Vicer versa...if 3 CCDs are needed for accurate colors, why have the professional digital still cameras like the Canon 1DS NOT incorporated 3 chips?

Just a thought!

Because still cameras do not need to take an image thirty times every second. Doing that HARD. Really hard. Even the fastest digital still camera (1d2m3) only does a third of that, and it is prone to overheat during consecutive use.

Amazingly, we have a large, single 1/3" sensor with excellent resolution, low noise, poor moire, but at a great price: $1,000 in an entire camera (I'm speaking of the Canon HV20).

If sensor design scaled up linearly, you might expect that triple the money ($3,000) would give you a sensor with triple the surface area. Instead, it merely gives you three of the same 1/3" sensors (i.e. the XH A1). That's because small sensors are easier and less expensive to produce than larger ones.

By the way, "good" color is subjective. Do you mean accurate? Or without noise? Moire? Single-sensor bayer, foveon, and prism-based methods have upsides and downsides, and noiseless, moireless, accurate colors can be achieved with each. Implementation has more of an effect on those metrics of color.

Different design for different purpose. Putting aside obvious reasons like back focus distance, crop-factor change, and loss of control over depth of field, there is a fundamental limitation that still cameras do not have to deal with: 30 frames per second. It makes all the difference.

Good point. Thanks Dan.

I overlooked the 30 times of data grab per second. Very true...none of the DSLRs has to do that!

And yes, a lot of people might have objection when they see me using the term accurate or precise color. I would like to support my argument by pointing out that digital files from pro DSLR imagers have far fewer inaccuracies in color and focus than the film. In other words, the output from DSLR imagers are much more consistent.

The only area that I'd give film a plus is gradation or continuity of tone. Of course, digital is discrete, and no matter how hard you try...8-bit 16-bit sRGB-Space ARGB-Space, you can never get really smooth color tone.

But I believe the end product by a DSLR -- say 1DSMkII -- is now so good to my eyes that I cannot find faults. Many museums and big-name magazines have accepted 1DS and its siblings as archival-grade equipment.

Perhaps we should discuss this more in a more appropriate category. This is Z1/FX1 topic. My bad. I talk too much and now have side-tracked too far off.

Cheers,
Alex

1- To get very good color accuracy, you want your sensor's spectral response to correspond to the viewer's cones' spectral response.

Where spectral response = sensitivity to each particular wavelength of light.

One set of commonly used measurements for humans' spectral response is the 1931 CIE figures... these give the CIE color matching functions (CIE CMF). To get a sensor with the same sensitivities as the CIE CMFs, you need 4 different colors/filters like the Sony RGB+E sensor used in some of their still cameras (though their implementation isn't 100% accurate?). Color calibration probes like Truelight's also use 4 different colors/filters.

2- Another reason why cameras aren't accurate is that they don't apply the color correction matrix the right way.

http://www.dvxuser.com/V6/showthread...ghlight=matrix

But in practice, you can get away with lots of color inaccuracy. (Real life is color inaccurate anyways / colors are a guess. It doesn't make sense for us to memorize subtle color variations.)

3- Camera/imager design:

3a- Some people like film because it has a bigger dynamic range than video. You can have hot zinging highlights that don't look weird when they clip.

Film's color response is very wacky. There are 3 or 4 color layers (depends on film stock), and the math gets complex because each layer blocks light from hitting the next layer and each layer's transfer curve is non-linear.

3b- For most (all?) 3-chip (including 3 CCD) designs, pixel shifting is used to oversample the signal. I don't know the exact details too much... but you generally want to oversample the signal so you can apply DSP to reduce aliasing and keep good sharpness. Pixel shifting improves luma performance at the expense of chroma/color performance.

The pixel shifted 3CCD design is also desireable since the pixels are larger and thus more sensitive.

If you want a shallow depth of field, then 3 chips give you a smaller imaging area than one bigger sensor.

3c- For digital cinema, the single-chip approach lets you use 35mm film lenses. The 3-chip design doesn't work with 35mm film lenses (you need a prism to break up the light, but it sets a minimum distance between lens and imager).

In other applications it might be a more efficient use of pixels (more resolution given the same # of pixels) / better bang for the buck. A single-chip design definitely does give you more resolution than a 3-chip design if there are the same # of total megapixels.

Single-chip designs can be prone to greater aliasing if there isn't enough optical low pass filtering or if you use a bad debayer algorithm. You see this in consumer cameras a lot.

3d- The Foveon approach (stacked sensors) has the disadvantage in that silicon is a poor color filter. Once you extract the color there is a lot of noise. (Search Graeme nattress' posts on reduser.net)

4- If you subscribe to the school of "just look at the damn results"... then the various imaging technologies have their pros and cons (except maybe Foveon, which arguably sucks in practice). And film delivers some of the best images around.

in fact there is no technical reason , why a solution working for photo would not work for video (and the contrary also).
this is probably just history.
People making still camera are coming from a place were the technology use big sensor (the film). they naturally move to digital by just replacing the film with ccd. no change in the lens, the reflex system, the case... etc.
On the other hand, movie camera for consumer was 8mm and super8mm or en 16mm (we will not take in account 35mm camera, since the guys using these camera are for most still using good old plain film).
These format were small film, then they were replaced by small sensor. Anthere was not a huge base of standard lens like in photography world.
I do not know if you heard before about the funny story about the space shuttle and size horse's ass.

http://www.astrodigital.org/space/stshorse.html

No. There are many technical reasons: semiconductor die yield rates, readout interference, and rate of hot and dead pixel occurance. Those are just the reasons obvious to someone who is *not* a sensor engineer; I'm sure there are more.

There is a huge market for large sensors for still and video alike. No one would take pains to maintain a historical limitation when riches are on the line.