Comparison of Canny operator and snake for segmentation
- A comparison was made between the results of canny operator and snake model for small regions, all from
110.pgm. These images are s1.pgm through s11.pgm. Results are compared and shown respectively
for the mentioned images. Prior to snake implentations, Sobel operator was used for gray-level image processing. Moreover, snake model
was applied with and without using GVF. For all images, except for s11.pgm snake failed in first implementation
iterations, which was due to the poor external force orientation for neightboring pixels with close gray scales. As mentioned in previous
reports, snake model using Sobel operator without GVF usually gives promising results. However, about one pixel shift from the actual
pisition of the edges is produced. Because the shift is totally appiled to the whole snake model, the results are still acceptable. Below, the original images are
illustrated by matlab code "rawread" to read image files with pgm extensions. Since these images are small, for a good comparison
with the canny and snake restuls, the matlab "rawread" versions are provided here. Thoguh through the hyper-text activated image names,
the actual original images are provided, they are blurred due to maximization. Also, the dimesions of the matlab read original images with their
actual type may be different; therefore, the matlab "rawread" version of original images were shown to keep the same dimensions, for
comparison with the snake and cany operator results.
- For s4.pgm snake shows a better fit since canny operator has streaking issue with mising edges:
- For s7.pgm and s8.pgmcanny operator has a preciser result than snake:
- For s10.pgm, which is tricky image, again snake shows a better
result than canny operator:
- s11.pgm is very tricky image. The region of interest is the dark area close to top surrounded
by gray boundary. Neither snake nor canny has a promising result, since canny has the streaking issue (missing edge points) and
doesn't detect the edge well at all, and for the snake, without GVF result is not satisfactroy and with GVF, snake model has a closer
fit but still not acceptable:
