[Leica] Re: do your homework, please!

[Alexey Merz <alexey@webcom.com>]

I response to a typically excellent Erwin post, 
Jim Brick wrote, about CCDs:

>The number of electrons held at the pixel site can 
>be from zero (no exposure) to I suspect
>billions.[...]
>So you see, digital cameras aren't really digital 
>at all. The analog light value collected in each 
>pixel is electronically "converted" to a digital
>value. The electron buckets (capacitor) in each pixel
>has to have a large enough range from 0 (black) to 
>billions (pure white) or the differences between 
>various shades of grey cannot be ascertained. 

Now we are getting somewhere; this is a good description.
But I should point out that Jim's a bit off on the numbers.
As I posted yesterday:

"The diodes in consumer-grade 2/3 inch CCDs usually range from 
about 7 to 13 um in size and hold a stored charge of 50,000-
150,000 electrons. A rough calculation of potential-well storage
capacity may be obtained from the area of a diode multiplied by
1000. For example, a diode of dimensions 10 x 10 um will have a 
full-well capacity of about 100,000 electrons."

(Inoue & Spring [1997] Video Microscopy: the fundamentals. 
2nd ed. New York: Plenum Press. ISBN 0-306-45531-5.)

So we are actually dealing with maybe 100,000 electrons per
pixel, not billions. And as we decrease the size of the pixel
things get worse: not only is the ceiling lowered, not only
does the total number of possible grey values decrease, but
as the wells get smaller, each grey value gets closer to zero
charge.

This means noise: noise from quantum effects within the well,
noise from ambient thermal effects (most CCDs are fairly 
infrared sensitive) and noise during the A/D conversion step 
because of smaller signal ampplitudes. All of these noise 
sources are significant, and all degrade dynamic range.

It's for this reason that scientific CCDs tend to have LARGER
pixel sizes of up to 25 x 25 um (that's 20 lp/mm in greyscale/ 
10 lp/mm in color, for those who are still counting). In 
scientific applications we often seek to minimize noise and 
maximiize dynamic range, and are willing to live with lower 
resolution. How do we reduce noise still further? Well, the 
CCD on the camera that I use is cooled to -50 C. Not a big 
deal if you are willing to live with a bulky & power-consumptive 
cooling system. But not real convenient for a camera the size
of an M6.

As Jim writes:
>This puts size constraints on the manufacturing of each 
>pixel. Make them smaller, the noise data goes up and
>the good image data goes down. You get more pixels
>per inch, but a crappy image.

He is not kidding, and those who blithely predict that CCDs
will soon *surpass* film in image quality simply don't 
understand the inherent constraints of current CCD imaging
systems.

- -Alexey