John Trinder
e Morphological tools for narrow features, such as roads on satellite images.
e To detect bright small elements, erosion followed by a dilation is applied; then the original image is subtracted
from this result.
e For dark small elements, dilation is applied first, then erosion.
3. Derivation of a chamfer image in which the pixel values relate to their closeness to any surrounding edge.
4. Acquisition of control points (initial curve) from the user that roughly delineates the feature
5. Smoothing the initial curve using B-spline and initialize its state
6. Minimize the energy of the state using SA
Figures. 22, ??, and ??, ??, demonstrate the performance of the snake and SA based method respectively for different
initial configurations. The white curves in the figures represent the B-spline drawn from the user provided control points
while the black lines represent the final solution. As can be seen from Figure. ?? the snake is able to provide a good
estimate of the feature of interest only when the initial approximation is quite close to it, but it becomes distorted by
suboptimal solutions otherwise (see Figure. ??). On the other hand, the SA is found to provide a good estimate of the
feature of interest even when the initial delineation is well away from it (see Figures. ?? and ??).
4 DISCUSSION
An efficient semi-automatic feature extraction technique which integrates the principles of active contour models with
simulated annealing is developed in this article. The power of the annealing procedure for providing stable minimum
energy configuration has been utilized along with the principle of active contours to search for appropriate feature de-
lineation which minimises the energy. The significant superiority of the developed technique over the snakes model is
demonstrated on an aerial image for several initial configurations. Note that snakes may become distorted by local en-
ergy minima, depending on the initial user delineation and repeated reinitialisations are required each time this happens.
Moreover, snakes are found to provide a good solution only when the initial user delineation is quite close to the feature
of interest. The developed technique does not suffer from either of these limitations, the only requirement being that the
window size is large enough so as to include the feature of interest. Note that the pull-in range of the SA based method
is governed by the size of the window. Therefore, increasing the size of the window will increase the pull-in range of the
method, at the cost of increase in its run time.
In this context, it may be mentioned that if more than one feature of interest is present in the window, then the SA based
method would capture points from all such features, thereby resulting in the loss of structural continuity in the final
solution. Additional terms may be incorporated in the energy function, which will automatically enforce the structural
continuity of the contour provided by the algorithm. Preliminary investigation in this regard are already being carried out
by the authors.
Acknowledgement : The authors acknowledge the Australian Research Council and LH-Systems for providing the grant
to carry out this research.
REFERENCES
Brown D. and Marin J., (1995). ‘Learning vector quantization for road extraction from digital imagery,’ Intelligent Systems
for the 21st Century, vol. 2, pp. 1478-1481, IEEE International Conference on Systems, Man and Cybernetics.
Canny J., (1986). ‘A computational approach to edge detection,’ IEEE Trans Pattern Analysis and Machine Intelligence
PAMI, vol. 8, pp. 679—698.
Fua P. and Leclerc Y., (1990). ‘Model driven edge detection,” Machine Vision Applications, vol. 3, pp. 45-56.
Gruen A. and Li H., (1994). ‘Semi-automatic road extraction by dynamic programming,’ International Archives of Pho-
togrammetry and Remote Sensing,, vol. 30-3, pp. 324-332.
Helmut M., Ivan L., Albert B., and Carsten S., (1997). ‘Automatic road extraction on multi-scale modeling, context and
snakes,” International Archives of Photogrammetry and Remote Sensing,, vol. 32, pp. 106-113.
908 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
