Edge detection: Canny - LOG comparison
- To apply snake algorithm on a gray scale image, an edge detection procedure is highly required to detect the
edge/boundaries. For instance, Snake modelit does not work successfully on the above-mentioned images. For that reason, two edge
detection methods are proposed (by Xu & Prince): Gradient and LOG. LOG was applied on "small.bmp/pgm" and different s=0, 0.1, 0.5,
1, 1.5, 2, 2.5 were applied and all failed.
- Canny detector similar to Sobel (gradient) operator was applied and the result are promising. It did not fail the way LOG
resulted. It was applied on 110.tif, small.pgm with the best conclusion in both of
them.Resutls:
- Some edge detection appraoches, using matlab default code "edge," were applied and compared on
110.tif. Results show a more promising edgedetction with canny operator:
- To find out what exactly canny detector is doing? If there is any thresholding, since the output of canny operator is
binary, try to exclude the thresholding. Why exactly LOG fails and what about gradient operators?
- Using "Edge" code of Matlab, LOG and Canny operator were compared for 110.tif and
small.pgm. Without regard the thresholding in that code which../ is also set automatically, different values
for s (Gaussian width) were assigned: 0.8, 1.0, and 1.5(results available for small.pgm:
.
There is a trade off within this operator since using small values of s
detected weak edges which are undesired, and larger values exclude some edge information which is required. The default value in
this code is s=2.
- For s=1.5, the snake result of small.pgm has some mismatch with the edge on the lower right side;
result of s=1.0 was similar to that of canny operator; and the result of s=0.8 failed on the boundaries of upper left side.
