# Last edited on 2012-01-16 01:44:57 by stolfilocal

HANDLING GRADIENT NOISE

  Tried hard to find a good approximation for the 
  distribution of the azimuth of (x,y) when 
  (x,y) follow a symmetric 2D Gaussian distribution 
  with mean (0,d) and deviation 1 in each coordinate.

  See compute-polar-gaussian.gawk to compute the true 
  distribution (by line integral along each ray).

  See plot-polar-gaussian.sh to plot the true distribution
  and some candidate approximation formulas.

  Eventually gave up and just reduced the weight of each 
  gradient by the factor {g^2/(g^2 + n^2)} where
  {g} is the gradient modulus and {n} is the specified
  noise deviation.
  
GENERATING THE TEST FILES

  Raw files created with gimp (centered between pixels):

    in/RAW/janus-big.png 400x400
    in/RAW/jdiag-big.png 400x400
    in/RAW/odots-big.png 400x400
    in/RAW/sqres-big.png 400x400
    in/RAW/trigs-big.png 400x400

    in/RAW/white-msk.png 400x400
    
  Raw files created with make_bullseye (centered between pixels):

    in/RAW/bueye-big.png 512x512
    
  Raw files created with make_test_slope_maps (centered at pixel)

    in/RAW/bdots-big.png 513x513
    in/RAW/bpent-big.png 513x513
    in/RAW/btria-big.png 513x513
    in/RAW/ptpyr-big.png 513x513
    in/RAW/wavec-big.png 513x513

  Reducing them to 100x100:
  
    make-input-images.sh