Archived posting to the Leica Users Group, 1999/12/10

[Author Prev] [Author Next] [Thread Prev] [Thread Next] [Author Index] [Topic Index] [Home] [Search]

Subject: Re: [Leica] scanner recommendation
From: Pitak Chenkosol <pitakc@ee.pdx.edu>
Date: Fri, 10 Dec 1999 14:15:18 -0800 (PST)

	Dear Tina,

	I am not a scanner expert in anyway but I will take a shot
	at this.

	What the manual tells you is basically the following:

	"The more times you scan the same image the higher signal
	to noise ratio you will receive as a final result."

	Ok, let me be more specific: 

	When scanning an image a scanner will send the scanned data to
	the host computer. This scanned data includes not only the actual
	image information but also system noise. This random fluctuation
	noise has a statistical property that its effective magnitude
	will decrease if you average (add them up and divide by the number
	of scan, for example) it. The term "random" is the key
	word here. You can use this averaging trick if the signal is
	random in nature. This is why the actual image information does
	not decrease with the averaging.

	Now let's look at a simple example. Assume that a scanned data
	from the scanner can be represented by

	2 + system noise.

	Now after you scan the same image 5 times, you will have the
	following:

	1. 2 + system noise (1st scan) 
	2. 2 + system noise (2nd scan)
	3. 2 + system noise (3rd scan) 
	4. 2 + system noise (4th scan) 
	5. 2 + system noise (5th scan) 

	If you now average the above results you will get :

	Average scanned data = (2+2+2+2+2)/5
			      +(system noise #1 + system noise #2 +
				system noise #3 + system noise #4 +
				system noise #5)/5

			     = 2 + effective system noise.


	The magnitude of the "effective system noise" now will be smaller
	than the individual system noise from different scans above.
	A comparison of the signal to noise ratio data will reveal that
	(2/effective system noise) is bigger than (2/system noise).

	This is what I meant when I said that the signal to noise ratio
	will be higher with multiple scan. Now you may say that that is
	great I can just scan my image a few more times and my image
	will look better. It sure will but you may not be able to tell
	the difference if the number of scan is already high enough.
	There is a practical limit.

	Now let's look at an extreme case. Say if I can manage to scan
	an image 1000 and 1005 times then compare the final results.
	It will be very difficult if not impossible to distinguish these
	results. This is a reason why you may run into a limitation that
	a scanner or scanning software company put a limit on the number  
	times you can scan an image. This is the practical limit. 

	Now that you have a higher value of signal to noise ratio from
	multi-scan. The image encoder (either hardware or software) will
	be able to provide you a result with higher bit depth (because
	there is more information to be encoded).

	The principle behind all this is the statiscal signal averaging
	in the presence of unwanted noise signal. It is a standard tool
	in signal processing. I put it in term of image scanning in order
	to stay on the topic of discussion.

	Hope this helps.

	Regards,

	Pitak


> 
> But the multi-sample scanning on the Nikon increases the bit-depth per 
> pixel by reducing system noise. "With a 4x scan the pixel depth is 
> increased by two bits per pixel and with a 16x scan the pixel depth is 
> increases by four bits per pixel."
> That is straight out of the manual.  I have no idea what it means, but with 
> contrasty slides the multiple scans greatly increase the shadow 
> detail.  It's not scanning the whole slide 16 different times, but scanning 
> each pixel 16 times as the scanner moves across the slide once.
> 
> Tina
> 
> 
> Tina Manley, ASMP
> http://www.tinamanley.com
> 


- -- 
Pitak Chenkosol, Dept. Electrical Eng.,|  " I was born not knowing and have
Portland State University,             |    only had a little time to change
P.O. Box 751, Portland, OR 97207-0751. |    that here and there." 
E-mail: pitakc@ee.pdx.edu              |                Richard P. Feynman