Synthetic images were created with a step edge (height S) and added Gaussian
noise (sigma N). Signal to noise ratio SNR is defined as the square of S/N,
or SNR = (S/N)^2. Using EdgeDetector 1.1, edges were
trial-and-error experiments with various combination of the parameters of LL and
ES. No satisfactory results were obtained from the SNR = 1 image.
Results from similar images using Sobel, East Compass, Roberts magnitude,
Roberts square root and Kirsch methods can be found on Pages 544 and 545 in the
book "Digital Image Processing" by William K. Pratt, second edition,
John Wiley & Sons, Inc., 1991.
You are welcome to download these synthetic images for your own experiments
using EdgeDetector or other methods. If you put
the results on your web page, please let me know so that I can put a link to your
page from this page for easy comparison.
More examples were available for real images.
The Gaussian-noise-added step edge images with SNR = 100, 50, 10, 5 and 1,
from left to right, top to bottom:
From the SNR = 10 image, the following result was obtained with two sets of
parameters: 1) LL=80 and ES=0.45; 2) LL=60 and ES = 0.98.
From the SNR = 5 image, the following results were obtained with 1) LL = 90
and ES = 0.7; and
2) LL = 90 and ES = 0.65.