in addition to segment image, also define a nonuniform mask
which represents all pixels which should not contribute to uniform image.

Need to make calculations using various uniform images to compare results.
Results need to include id of uniform images used for calculation.

Need to be able to cycle through a series of these images (on disk)
to produce one uniform image from original 16 bit data and the nonuniform mask


5/4/94 notes

How to combine bright images:

1.  subtract dark from each one.
Pick one image to be denominator image.
2.  take ratio of a pair, then make histogram of the ratio image,
then find the median from histogram.  Equal values give 32768 (1.0).
3.  repeat this for each image against one image.
4.  multiply all images by value such that median of ratio will be 32768 (1.0)
Now the images are comparable and can be combined by averaging or median of images.

end 5/4/94 notes

procedure for making a uniform image:  

convert to 8 bits.
move a square roi over image.  
measure mean and stdev of roi.
copy inset square to output image.
set threshold values to one sigma below and above mean, make binary.
move roi

invert resulting binary image and use as mask for masked convolution


locate pixels which are within one sigma of mean of a fairly large square roi.  



OR:  finds the median of 9 images,
then reduce noise and smooth, 
then run flat field correction on each image,
threshold it to within 1 sigma of mean and use masked convolution to
find a smoothed uniform image derived from the original raw pixels
corresponding to the one sigma pixels of flattened image.
Then run median again.