RE: [APML] Fw: images

["Lane, Jason R" <jason.lane@navy.mil>]

Hi Chuck,

Thanks for the replies.  I've found that on my slide scanner (granted it's just an HP S10xi), scanner noise is an issue.  I have to compensate for this by averaging several scans.  I suppose more modern scanners do the same thing by making multiple passes at a slide.  

Being a ratio, S/N is a unitless number.  Take the bit value (or irradiance level or whatever) of your target and divide it by the bit value (or irradiance level or whatever) of the background.  Whether you use absolute, RMS, or average values depends on who you are trying to impress I suppose.  

I am not sure that visual inspection of differences between prints or displays using different numbers of levels is a good way to compare information loss.  Granted, for recreational use it's fine.  However, reducing the amount of data in a scan by any amount (going from, say 16 bit to 15 bit) will result in the loss of information.  For scientific use, this could be a critical difference...especially if you were dealing with quantitative data...that is looking for actual values vs. comparative.  I can say as an engineer that bit depth would bug the heck outta me, much like a guy who restores cars grits his teeth at "20 footers"..that is, cars that look good from 20 feet away but have numerous dings and flaws when you get up close.

The other problem with visual inspection is that the results will vary from one individual to another.

Regardless, for the example of making a print...it seems intuitive to me that you would want to select a greater numbers of levels for the scanned image (bit depth) when sending the image to the printer.  Wherever the color space conversion is made (whether manually in Photoshop or automatically on the way / in the printer), the minimum number of bits required to view the image on the screen would limit the ability of the conversion process to provide you with a print that looks the same as the original.    

It also seems intuitive that for processing reasons you would want to record and store the image at the highest bit depth possible...this gives you the greatest flexibility when processing the image and making cosmetic adjustments.  Very likely you can store the final result at a much reduced bit depth, but it doesn't seem necessary unless your hard drive is extremely small.

Nevertheless, your statement seems reasonable for everyday use...although I agree with you that the nomenclature can be confusing.   It just shows the lower level of standardization in the Optical community compared to, say the Electrical Engineering field or even compared to Mechanical Engineering.

-Jason

-----Original Message-----
From: astro-photo-bounces@seds.org
[mailto:astro-photo-bounces@seds.org]On Behalf Of Chuck Vaughn
Sent: Wednesday, June 23, 2004 6:54 PM
To: Discussion of Film Astrophotography
Subject: Re: [APML] Fw: images


Jason,

> I'm not sure how useful it is to talk about bit depth with an analog 
> media, since the amount of exposure is not recorded in digitized steps 
> (or histogram bins) in the silver halide grain.  Are you saying 5-6 
> bits of information after conversion to digital?

Yes. The noise-free range can be represented by 5-6 bits of data.

> Do your numbers account for scanner noise?

Scanner noise is not really an issue unless you're close to the Dmax 
the scanner can handle. I have trouble with scanner noise on Kodachrome 
slides and dense TP negatives.

>  6 bits =  64 levels per channel, which is hard to believe for 
> something like a color photograph.

64 levels per channel would give you 262,144 colors. You can't see the 
difference between that and 16.7 million on a monitor or a print. Try 
it for yourself in PS.

I have noticed that monitors don't have enough dynamic range to see 
anymore than about 128 gray levels. I could see no difference at all 
between 128 and 256 gray levels.

>   It is awfully similar to what practical tests have shown for S/N of 
> film:  A S/N ratio of 4-5 if the element is to be detected against its 
> surround, 8-10 if it is to be recognized.

What units is this in? Probably something I never heard of. :-)

>   I don't know if it's correct to talk about film directly using 
> digital nomenclature...maybe for a scanned image of a film negative, 
> though.  I think this requires some clarification, because even after 
> dealing with this stuff at work I must admit you have me confused.
>
> On the other hand, there is an expression giving the information 
> capacity of film..that is the bits per unit area that can be stored on 
> a photographic layer.  Altman and Zweig (Phot. Sci. Eng. 7:173, 1963) 
> reported levels up to 160x10^6 bits/cm^2 for one set of fairly 
> stringent conditions.

We may have some apples and oranges comparisons going on here. Remember 
I'm talking about astrophotos. On color film the images are mostly very 
thin so that reduces the S/N. They'd be just as thin on TP if we didn't 
develop it in D19 to really enhance the contrast. Again that reduces 
the S/N. TP developed in Technidol is another story.

All I'm saying is that the signal levels we typical deal with on film 
astrophotos can be represented by 5-6 bits when converted to digital.

>  Naturally, an actual film would have a gaussian-type distribution of 
> actual # of grains at a given "bit depth" (due to grain size, etc), 
> but it probably is a good estimate.

I played around with this a little in PS, starting with a 60% gray 
image. I added various amounts guassian noise. This is just an 
observation, but 2 times the standard deviation converted to bits seems 
to give an estimate as to how many bits are noise.

Chuck

_______________________________________________
Astro-Photo mailing list
Astro-Photo@seds.org
http://seds.org/mailman/listinfo/astro-photo
_______________________________________________
Astro-Photo mailing list
Astro-Photo@seds.org
http://seds.org/mailman/listinfo/astro-photo