Has CCD image sensor technology improved?

One notices manufacturers are offering higher megapixel cameras with
the same size CCD image sensors. Also, the recent Nikon D70 comes with
a CCD sensor, while most other digital SLRs hav CMOS sensors.

Has there been any dramatic improvement in CCD image sensor technology
that allows higher density of pixels without increasing power
requirement?

Who makes CCD image sensors? Who makes them best?

How much does colour array system (CYGM, RGB, et al)make a difference
in the pixel density/ noise rendition?

Shall appreciate any informed answers.

Thanks
JG

 

Check out Foveon theory, good Idea good pictures,but only one manufacture
uses it. Time will tell. Lower noise than the Point & shoot cameras but not
up to par with the Canon cmos.

 

Pixel density and power are not directly related.

CCD technology is very mature and really has very few places to go.
larger sensors and/ or more pixels is really tied to Silicon
processing technology and yield

MANY people.

99% of all digicam sensors are Bayer filter devices using Red, green
and blue filters. Sony has added the "emerald" filter---claiming
improved color rendition.

AFAIK, no one uses CMYK in sensors.

I dont know what "CYGM" is (typo??)

The filter pattern, color space, etc. is independent of noise and
pixel density

Also take a look at the tutorials on sites such as dpreview
**************************
Mark Herring, Pasadena, Calif.
Private e-mail: Just say no to "No".

 

Can you explain this please Mark, or point to an explanation.

Thanks.

 

Not really---read some of the 1000s of posts here

Foveon, IMHO, has a fundamental flaw: You cannot control the spectral
bandpasses for the RGB channels. Thus, you cannot get as good a clor
balance as with the Bayer

BTW, is not the Foveon a CMOS (not CCD) device? I've forgotten.
**************************
Mark Herring, Pasadena, Calif.
Private e-mail: Just say no to "No".

 

There is some stuff on this in this NG. search in Google Groups

Here's my 2 cents:

For any color imaging system, one needs to approximate the spectral
response of the human eye. In fact, the whole concept of "color" is
tied to our eye-brain process. With film, dyes are individually
tailored to try and get the right balance. With the Bayer filter
pattern, you have individual filters whose bandpasses can also be
tailored.

With the Foveon technology, the spectral properties are determined by
the absorption depth in Silicon. This is an intrinsic property and
does not provide the flexibility to individually control the R, G, and
B bandpasses.

Many people have reported color accuracy issues with the Foveon.
Above is only my hunch as to what at least part of the reason may be.

-Mark

**************************
Mark Herring, Pasadena, Calif.
Private e-mail: Just say no to "No".

 

I don't see why that would make the colour accuracy any worse in itself. As
long as you can get consistent depth of silicon across the chip through the
3 layers, it shouldn't be a problem. I don't know alot about silicon fabs
though so I've no idea if this is difficult to manufacture or not.

Ahh a hunch. I thought you'd stated it as more than that.

Thanks.

 

There are two separate points here.

Firstly, as I understand it, there is a problem with absorbtion in silicon.
It is much easier to manufacture a filter of the correct colour than to
manufacture sufficiently accurate silicon 'filters'. Foveon is a neat idea
though.

Secondly, as Bayer takes account of our perception of colour, Bayer
effectively records luminance/resolution at the expense of accurate
colour in every pixel. Exactly the same principle is used in the television
and in our eyes - we see detail better than we see colour.

Graham

 

Secondly, as Bayer takes account of our perception of colour, Bayer

But in colour TV, great effort is taken to ensure that the value of the
colour is correct, the vector part of the chroma. The in-accuracy is in
the precise spatial location of the colour, not in its value.

By contrast, the silicon filters appear to have two problems: (a) the
three sensor values cannot be accurately dematrixed to RGB space and (b)
the filters vary in response over the area of the sensor. (b) could
presumably be solved by the appropriate post-processing, but (a) seems a
more fundamental flaw right now.

Cheers,
David

 

Consistency is not the main problem with the Foveon approach. The biggest
problem is discrimination. They have chosen a material (doped silicon)
primarilly for its electronic properties (i.e. they need to fabricate a
working integrated circuit from it) and not its optical transmission
properties. In prcatice, it appears that they have difficulties, compared to
Mosaic sensors, in sampling a sufficiently narrow spectral band in each
channel.

In other words, their colour samples are unsaturated and not well seperated
according to frequency. In order to get good, saturated colour from these
samples they have to resort to subtracting from each channel, weighted
versions of the other two during processing. This results in increased noise
in their output, as well as poor colour reproduction, especially around
saturated highlights, where visible colour-banding/posterisation is often
visible in its output.

You don't have to take my word for this, you can examine the raw,
unprocessed output yourself. Just take Dave Coffin's dcraw source code and
remove the calls to foveon_interpolate() and foveon_coeff(). This will give
you a .psd file which contains the unmanipulated samples from the sensor.

I did some very rudimetary investigation of this myself. The results are here:

http://homepage.ntlworld.com/narcissus/SD9/

It's possible I might have got better results if I'd had a specific profile
for the SD9, but even then, it's pretty clear that the numerical pixel
values for greens and blues are very similar, as they are for reds, yellows
and purples. In particluar, highly saturated greens are almost recorded as
grey in the raw sensor data, and the "blue" channel is very dark, which
probably explains the blue-sky noise which can characterise SD9 images.