Help for REMNOISE

PURPOSE:
Remnoise removes small artifacts, like bit errors, from images.
At low error rates it performs the same as ADESPIKE. Performance is superior
to ADESPIKE at bit error rates over one in 500.

EXECUTION:
remnoise inp=image out=cleaned tol=30.0

Note:
The TOL parameter is somewhat sensitive to bit error rate and very sensitive to
the image itself. To tune it take a typical image of interest without noise
and run the following test with RATE=the expected bit error rate:

addnoise inp=image out=a.img gain=30 rate=100
remnoise inp=a.img out=b.img tol=10 factor=.5
f2 inp=(image,b.img) out=c.img function="in1-in2+100"
hist inp=c.img 'nohist

repeatedly changing TOL each time until the standard deviation listed by hist is 
a minimum.

The FACTOR parameter is insensitive to bit error rate.

METHOD:
1. A 3 by 3 pixel box is centered on each pixel of dn J.
2. A median is computed taking each consecutive 3 pixels going around the box.
   This makes 8 median values I.
   I1 I2 I3
   I4    I5
   I6 I7 I8
3. A surface is interpolated using I1-I8
   The model is: M=sum(I/R)/sum(1/R) where R is the distance of I from the
   center (1 or 1.414).
4. An activity measure A is computed from:
   A=(|I1-I8|+|I3-I6|+|I2-I7|+|I4-I5|)/4
5. If abs(J-M) > TOL+FACTOR*A then  J=M

HISTORY:
1-1-99  J Lorre. 
COGNIZANT PROGRAMMER:  Jean Lorre

PARAMETERS: