next up previous
Next: Two Dimensional Features Up: Edges Previous: Local Edge Context

Other Methods

Finally, some edge finding methods which do not fit any of the above categories are briefly described, for the sake of completeness of the review.

In [91] Shann and Oakley implement Canny's edge finder and the LoG filter in a new way. Instead of convolving the discrete image with a discrete approximation of an edge enhancing filter, they use a (virtual) continuous convolution filter to give a filtered virtual image which is continuous; this is analyzed using numerical methods to find exact edge positions. No results are presented.

In [34] Eichel and Delp use a small scale local operator to measure image variance to enhance edges. Results presented are poor.

In [108] Strickland et. al. assume that the edge profile and the lighting conditions are fixed, and use a correlation-based template matching scheme to find edges. Within the limitations of the assumptions, the results presented are very good. However, the method clearly does not have general applicability.

In [27] Dattatreya and Kanal use one dimensional Kalman techniques to find smooth edge contours. Potential edge points are tracked in the x and y directions independently using Kalman filters. The two sets of results are combined to give an edge map. As with the context dependent approach of Haralick and Lee, it is not clear how junctions would be handled, and no results are presented of complicated images.



next up previous
Next: Two Dimensional Features Up: Edges Previous: Local Edge Context



© 1997 Stephen M Smith. LaTeX2HTML conversion by Steve Smith (steve@fmrib.ox.ac.uk)