What are accepted figures for signal/noise (SNR) and dynamic rangefor CRT? LCD? Film? Human eye?

I'm converting my old analog scanning electron microscopes to digital/pc
based.

One of the important questions is how many bits of resolution ie: levels
of greyscale are required. I've heard the analog front ends of the
scopes have a maximum SNR of about 72 dB, or 12 effective bits, and to
give me some flexibility for digital image processing that's about what
I'll shoot for: a bit better if I can get it.

But on the other end of the scale, I'm wondering what are accepted
figures for signal to noise or dynamic range of various display
technologies, and of the human eye.

It seems to me the published "contrast" figures for LCD displays of
about 600:1 peg them at about 49dB dynamic range.

What is generally accepted for B&W film? Is it highly dependent on film
type and processing? Anyone know some typical figures for various
processes? Is "silver print," which in my mind is marked by high
contrast, a special technique or just a fancy word for black and white?

How about for CRT's? Deos it depend in part on the design and
construction of the CRT, and do dedicated monochrome CRTs have a greater
dynamic range for greyscale than color CRT's used to display greyscale
images?

And what about the human eye, for that matter? It's been suggested the
human eye has only about 6 bits or 36 dB dynamic range/SNR for greyscale
images. Is that about right? Bonus question: I know the sensitivity of
the human eye varies with color, being most sensitive at about 555 nM
green. How about dynamic range and SNR? Deos it vary with color, too?

Thanks!

-Jeff

 

Most of what you ask I don't know, but what I _do_ know is this:

In infrared imaging (which is what I'm familiar with) you take as many
bits as you can get at the sampling rates you need -- this is currently
14 bit ADC's at 10MHz on high-end systems. Even so you need some
up-front gain selection (in the form of allowing the user to select a
few different integration times) to get the best picture for the
conditions. This data gets processed, gained up linearly or
non-linearly, possibly spatially and/or temporally filtered, then
presented to the user on video with an 8-bit DAC. Needless to say you
throw away a _lot_ of information when you go through the
user-presentation layer.

 

I am not certain about the technical data, you should ask some electronic
engineer, but it's very difficult to construct a lcd or tft scrren that can
compete a good crt, because the picture in the crt is made by tiny bits of
phosphorus struck by the cathod rays, also emitting their own light.Besides
that, crts have a developing history of more than 50 years.

 

Man! you ask a lot of good questions! I suspect that many of the answers
are to be found in papers of the SMPTE.

I recall that the contrast between printer's ink and glossy paper is
about 10:1 and B&W prints on glossy paper are a bit better (but on
matte, a bit worse). The contrast ratio of newsprint can be as low as
3:1 before the ink starts to look gray. Discouraging, no?

Look at your TV screen when the set is off. No part of it gets darker
when the set is on, but it sure looks like it does.

Jerry

 

Jerry Avins wrote:

...

Correction: 30:1.

Typical newsprint before the now-common biodegradable soy ink was 10:1.

Jerry

 

Photographic paper can get to 2.0 od reflected, a range of 10 ^ 2 = 100:1.

I just took a measurement from a printed page on coated stock, it
yielded 1.27 => 1.3 => 10 ^ 1.3 => 20:1.

 

Thanks for the reality check. The 30:1 figure was the best available,
typically found in (new) eye charts and photographic resolution targets
when I played with that stuff 40 years ago.

Jerry

 

The eye is comfortable with a brightness range of ~50-100x in one scene
with no pupilary/retinal adaptation, 1000:1 is viewable without noticing
much adaptation - after images or time for pupil to adjust. The eye can
accommodate a brightness range from reflected starlight on a field
(remember clean air and no streetlights?) to sun on fresh snow (except
in extreme cases, mountains, clean air ...). This about a 25 bit/zone
range or ~32,000,000:1 and has no relevance to viewing a CRT.

Within a 2.0 OD range it can easily discriminate to .005 OD in the middle
of the range. However, OD is logarithmic and photocells and CRTs aren't.
Figure you will need about 12 bits for very good fidelity, but 8 bits
looks just fine on a CRT.

Numbers above are from memory, you should check with a good text on
human visual response if the values are important.

Negatives have close to no relation to photoptic response. Exposure
and development of film are such as to yield the thinnest (least dense)
negative that will yield a good print. This figure depends on scene
contrast, film contrast, paper contrast and the method used to image the
negative on the film.

I have to confess I don't see much relevance in all this to an SEM. All
you want is the range of signal that presently goes to the current
analog display and be able to accommodate that. 8 bits resolution should
be adequate.

 

In normal lighting, the cones on the retina can detect contrasts of aprox.
2% over a range of 2 orders of magnitude. Their dynamic range is closer
to 4 orders of magnitude, but their ability to differentiate shades drops
rapidly at the lower (darker) end of the scale.

A 13 or 14 bit (linear) number would cover the same range, with the same
(or better) ability to separate shades. (Aside: Computer displays manage
with 8 bits, partly because they don't have a contrast ratio of 1:10^4 and
partly because they are not linear.)

There's a lot of good stuff at
http://webvision.med.utah.edu/
and
http://faculty-web.at.northwestern.edu/med/fukui/Human eye.pdf

The response of B&W film covers about 3 orders of magnitude. For B&W
paper, the reflectivity ratio between white and the deepest black is about
100:1. Ilford publish a lot of useful data on photographic materials at
www.ilford.com

-Tim

 

Nicholas O. Lindan wrote:

...

Yes. There's more light reflected from a lump of coal in sunlight than
from a bowl of snow indoors. The eye/brain isn't fooled.

For high-resolution work, we used to dye the emulsion black to avoid
exposing interior grains. After all, the image can't be in focus
throughout the emulsion depth. To make the film a little faster, we
would etch away the top few microns of the emulsion, leaving an exposed
layer of grains, like in sandpaper.

The question of which 8 bits can be important. I can adjust images from
my 12-bit flat-bed scanner in ways that just don't apply to 8 bits.

Jerry

 

But that is one of the attractive features of this group.
( OK, so I'm reading comp.dsp

By reading one learns so much more than one narrow area of knowledge!

 

I think that touches on a question I raised in another subthread.

Why a *FAST* *16 bit a/d* ?

Would it not be more suitable to have 2 A/Ds in "parallel"?

A very fast 6-8 bit A/D for use while positioning sample.
A 16+ bit A/D for analytical output.

Caveat Lector ;}

 

SNIP

See e.g.
<http://web.mit.edu/persci/people/adelson/checkershadow_illusion.html>

Bart

 

As others have said, you should capture and manipulate your original images
at the highest bit depth you can handle, but what is usually transmitted by
TV companies, recorded on DVDs, or saved in JPG files for display on CRTs
is based on 8 bit linear coding after gamma correction, and it generally
looks just fine. i.e. 8 bits seems to be quite sufficient for display.

From personal observation, dynamic range of monochrome and colour CRTs are
about the same, though the detail resolution can be slightly higher on a
monochrome CRT displaying a monochrome signal.

Rod.

 

Hmmm.... I'd thought about that but not to those extremes: in fact it
makes much more sense at the extremes you suggest and I may well
incorporate the idea. It will also give me a chance to jump-start the
project before I have the funds for the higher performance A/D.

Interestingly enough the 16 bit 5MSPS sigma-delta A/D I was thinking
about has a 10MSPS "turbo" mode with 14 bit resolution.

This thread is providing some very useful information about SNR of media
and human eyes... keep it coming!

 

Yes good point as there is no "dot pitch" issue.

-Jeff

 

See:
Notes on the Resolution of the Human Eye
How many megapixels equivalent does the eye have?
The Sensitivity of the Human Eye (ISO Equivalent)
The Dynamic Range of the Eye
The Focal Length of the Eye:
http://www.clarkvision.com/imagedetail/eye-resolution.html

Contrast thresholds data:
http://www.clarkvision.com/visastro/omva1/index.html

Roger

 

Great resource. Based on that data, I'd conclude that images captured
from my microscope might benefit for 65K x 65K resolution, so I'll use
16 bit D/A's to drive my scanning coils. Of course that kind or
resolution can only be rendered in print. And represents an 8GB file
size. And will take 2 hours to collect.... but might as well build it in.

The figure of 10,000:1 dynamic range in any "one view" corresponds to
about 12 bits. That's pretty much exactly the nominal dynamic range I've
heard quoted for the front end of an SEM.

But it might seems no current reproduction technology acheives quite that.

-Jeff

 

Hi Jeff
If you have a set of 40 cards with gradations of gray going from black
to white about 90% of the population will be able to lay them on a table
sorted into the correct order. If you increase the number of cards to 64
about 2/3 will be able to sort them correctly. If you up the number to
80 the number of people that can sort them drops to something like 20%
and by the time you get to 128 cards almost nobody can put them in the
right order. I'm recalling this from memory but the numbers are
something like that.
The sensitivity to brightness for film and the human eye is not
strictly linear. That is even when you give people 128 cards with linear
gradations of grayscale cards to sort there are certain ranges of
brightness wherein they will mostly get it right. So 8 bits (256
brightness levels) will pretty much handle the non-linearty of both the
display and the viewer. That is, most people will not be able to see the
difference between any two adjacent gray levels.

If your next question is - Is it spelled "grey" or "gray"? I don't know
:}
-jim

 

jim wrote:

...

...

Both are used, but "grey" is used less. The disappearance of "gaol" in
favor of "jail" is now nearly complete

Jerry