Digital camera images and What can you see

[Peter Pediaditakis]

In article <3AAE7A3B.7A55386F at unav.es>,
 jjmirujo at unav.es (" J. Martinez-Irujo") wrote:

> WSchick at aol.com wrote:
> 
> > In a message dated 3/13/01 1:39:22 AM Pacific Standard
> > Time, jjmirujo at unav.es writes:
> >
> >> And remember than most systems work with visible light,
> >> so do not expect to
> >> find any difference  your eye do not see...
> >
> > On both scanners and digital cameras, you CAN find
> > significant differences
> > that your eye cannot see.  Our eyes are optimized for
> > color, not B&W, so the
> > general level of grayscale differences our eyes can see is
> > about 30-40
> > grayscales, or a little over 1 OD.

I do not understand what you mean by "30-40 grayscales", could you 
please elaborate?

I am hard pressed to believe that significant differences could exist on 
a image that the eye could not detect.  Possibly on the low end, but if 
it (the pixel grayscale value) is close to background, one would have a 
difficult time defnding the results.

> >
> > Film such as a Polaroid or X-Ray is linear to about 80
> > grayscales, maybe you
> > can calibrate to 100  which give you 1.8 to 2.0 OD.  And
> > of course, your eye
> > can still see only 30-40 grayscales on the film.  Cameras
> > and scanners begin
> > at 8bit imaging (256 grayscales) and some are 12 bit or
> > even 16 bits.

Anything more than 8 bit depth is completely and utterly worthless.  The 
differeces of the 7th bit on is nothing more than "noise".  This can be 
shown by simply taking a fresh piece of film and developing it without 
any sample exposure.  Scan this and you will see that there is quite a 
bit of variability in the from the 7th bit on with regard to pixel depth.
 


> >
> 
> I agree that these devices are able to precisely measure the
> optical density of the film. In our lab we calibrate the
> scanner using a commercial photographic step tablet (density
> range approximately 0.05 to 3). By eye and putting the film
> on a white paper (reflectance mode )  you can see
> differences in the 0 –1 O.D. range. If you  put the film
> towards the light source (transmittance mode) you can see
> differences up to 1.7 O.D. This density corresponds to an
> overexposed film (by eye). By using the scanner it is
> possible to extend this range and see differences above and
> below this level that may not be evident to the eye (in this
> case I prefer to put another film to get  the right
> exposition). Within the 0.05-1.7 O.D. range, however,
> significant differences must be evident both before (by eye)
> and after image analysis, unless very small differences are
> expected (30-40 grayscales correspond to 2.5-3% difference
> in intensity).
> 
> While image analysis will provide an accurate measure of
> density in the film, usually this is not the critical factor
> in quantitation. If you run an internal standard, for
> example several dilutions of a radioactive sample, and plot
> O.D. vs radiactivity, you usually will find a linear
> relation in a range. Below or above this range the
> proportion does not apply. Moreover, in many cases the
> extrapolated line does not cross the x=0, y=0 origin, so
> doubling the amount of sample does not result in doubling
> the measured O.D.

First of all, unless nothing is there, you would expect the calibration 
line to not run through the (0,0) point.  Second of all, usinfg the OD 
is a mistake.  One pixel does not make a band.  In order to properly 
extrapolate, you must find the integrated optical density (IOD) over the 
range of pixels that constitutes your band.
 
> There may be many explanations for this
> behaviour and this is the reason that internal standards
> must be present. In fact, is the user, not the software, who
> by chosing a variety of user-selectable parameters that
> affect the quantitation (background subtraction, band
> intensity, band contour...) takes the responsibility of the
> result.

The only user selectable parameter that is of importance is the contrast 
used during scanning. Background subtraction should happen to all pixels 
on the scan uniformly.  Band intensity is what one is trying to find.  
Band contour,   I have no idea what that is maybe you could elaborate.

-- 
Peter Pediaditakis
University of Pittsburgh
Dept. of Pathology

remove "d-l-t" for reply