The required segmentation depends on the practical application. The chamfer matching algorithm only requires that image F be segmented to a binary image as input to the distance transform, and that a set of contour points be derived from image G. The contour points may be derived using a simple contour tracer, or by scanning a binary image for all nonzero pixels (the order of the points in the drawing is irrelevant). The algorithm imposes no constraints on the applied segmentation algorithms, other than that the number of points in drawing r must be relatively small for efficiency reasons. In practice this means that the point list is reduced by resampling. Later we will show that the chamfer matching algorithm is highly robust for the quality of the segmentation. This feature allows the use of "poor quality" automatic segmentation algorithms, i.e., the algorithm can be fully automated even though the quality of the images is poor.
Was this article helpful?