Page 258 -

246 6 Structural Pattern Recognition

Figure 6.2. (a) Octal primitives (top) and templates (bottom); (b) Binary image
with contour line segments according to the octal primitives.

Curve segments

Images described by curve segments, particularly line segments, can be obtained
by applying edge detectors to them, as shown in Figure 6.3. They are usually called
shape primitives or image silhouette primitives. In some applications, instead of
image silhouettes, one may be interested in obtaining a skelotonised description of
an image, by applying thinning or medial axis transformation to a binary image.
An application of this technique can be found in character recognition (see e.g., Di
Baja, 1996).
After obtaining the curve segment description, one may label the segments in a
similar way to what we have done for line primitives of signals, or extract other
descriptions from the curve segments (perimeter, curvature, slope of the segment
linking the ends, etc.).

Figure 6.3. (a) Image of a tank (toy model); (b) Image contours detected by an
edge detection algorithm.