ation is 
a set of 
b(i) are 
sonable 
ues that 
, which 
f c(1). 
ion and 
t shape 
hniques 
. These 
leriving 
or two- 
plied to 
noment 
that are 
nts that 
(central 
ult of a 
f seven 
and its 
ie shape 
applied. 
x and y 
> of the 
rotation 
ation is 
. Chain- 
basis of 
) and to 
»rmal to 
in code 
we eight 
| ‘7’ and 
tiples of 
oundary 
rding all 
he work 
he local 
  
Laura Keyes 
maximum gradient magnitude and that the gradient direction indicates, approximately at least, the neighbouring 
boundary pixel. More formally, the boundary direction D of an edge point is given by the edge gradient G as D = G + 2 
for the forward direction and as D = G -2 for the reverse direction (+90° and -90° respectively). By tracing the 
boundary, candidates for inclusion in the boundary are given by directions, D, D + 1, and D — 1. These candidates are, 
the pixel directly ahead of the current pixel and one pixel either side of it, for example if D = 1 (boundary direction is 
45°) then pixels 0, 1 and 2 are chosen as candidates. The potential of each of these candidates to be a boundary point is 
evaluated. If the pixel has been visited previously or if the pixel overlaps the image boundary, then it is assumed to have 
a negative potential. The candidate with the highest positive potential is selected as the next boundary point from which 
to continue the trace. This point is implicitly included in the list of boundary points by updating the BCC. 
  
Lind 
Pe 
  
  
  
  
  
(a) (b) 
Figure 1 (a) shows the eight neighbours numbered ‘0’ through '7', counter clockwise. (b), indicates the direction of 
each link (the direction that can be travelled between a point and its neighbour). 
BCC:31111757555 
  
Figure 2: A Boundary Chain Code (BCC) representation of a simple shape. 
This BCC technique outlined above is dealing specifically with digitised, edge detected pixel images. However, for the 
purpose of this paper this form of shape description needs to be adapted to deal with the vector map data described in 
the beginning of section 2, which will be seen in later in the paper. Given the BCC representation of a corpus of shapes 
in what way can they be classified. While the BCC is a useful method for the representation of shapes, recognition is 
normally based upon other descriptors derived from the BCC. For example, the moment shape descriptors discussed 
previously can be generated from the boundary points given by the BCC, which is utilised for this project. 
Scalar descriptors are based on scalar features derived from the boundary or an object. They use numerous aspects of 
the object for performing shape recognition. Simple examples of such features include: 
e the perimeter length; 
the area of the shape; 
the ratio of the area of a shape to the square of the length of its perimeter (A/P*); 
e the number of nodes; 
e the number of corners; 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 483