 
 
 
 
 
 
 
  
![[*]](http://vismod.www.media.mit.edu/vismod/support/latex2html-98//cross_ref_motif.gif) ) are then
considered as possible contributors to a template match. The magnitude of each
edge determines the extent of its contribution. Furthermore, the orientation
of the edge will also vary the contribution. If an edge is parallel to the
boundary of the template, it is well aligned with the template's contour and
it is a component of a boundary that is part of the overall shape we are seeking.
However, if an edge is perpendicular to the boundary of the template, it is
probably part of an external contour that crosses through the template. Such
an edge's contribution should therefore be attenuated. In other words, edges
which are not perpendicular to the normals at the template's boundary are
misaligned and their magnitude should be weakened to reflect this. The normals
of the template in Figure
) are then
considered as possible contributors to a template match. The magnitude of each
edge determines the extent of its contribution. Furthermore, the orientation
of the edge will also vary the contribution. If an edge is parallel to the
boundary of the template, it is well aligned with the template's contour and
it is a component of a boundary that is part of the overall shape we are seeking.
However, if an edge is perpendicular to the boundary of the template, it is
probably part of an external contour that crosses through the template. Such
an edge's contribution should therefore be attenuated. In other words, edges
which are not perpendicular to the normals at the template's boundary are
misaligned and their magnitude should be weakened to reflect this. The normals
of the template in Figure ![[*]](http://vismod.www.media.mit.edu/vismod/support/latex2html-98//cross_ref_motif.gif) are displayed.
 are displayed.
Observe Figure ![[*]](http://vismod.www.media.mit.edu/vismod/support/latex2html-98//cross_ref_motif.gif) . On the left is an annular sampling
region which is triggered by 4 edge segments:
. On the left is an annular sampling
region which is triggered by 4 edge segments:
 and
and 
 .
However, the edge
labelled
.
However, the edge
labelled 
 is not aligned with the shape of the template. Its
normal is not pointing radially inward/outward from the center of
cocircularity as is the case for the other edges. Thus, it is misaligned with
the radial normals typical of this circularly shaped region. It's contribution
is thus weakened by scaling down its magnitude so that it is equivalent to a
weaker yet properly aligned edge. Turning our attention to the ring on the
right of Figure
is not aligned with the shape of the template. Its
normal is not pointing radially inward/outward from the center of
cocircularity as is the case for the other edges. Thus, it is misaligned with
the radial normals typical of this circularly shaped region. It's contribution
is thus weakened by scaling down its magnitude so that it is equivalent to a
weaker yet properly aligned edge. Turning our attention to the ring on the
right of Figure ![[*]](http://vismod.www.media.mit.edu/vismod/support/latex2html-98//cross_ref_motif.gif) we can see
 we can see 
 now in a
lighter shade of gray, depicting the attenuation its magnitude has undergone.
Its contribution to the template matching algorithm is weakened yet not
totally discarded.
now in a
lighter shade of gray, depicting the attenuation its magnitude has undergone.
Its contribution to the template matching algorithm is weakened yet not
totally discarded.
Equation ![[*]](http://vismod.www.media.mit.edu/vismod/support/latex2html-98//cross_ref_motif.gif) is used to scale the magnitude of each edge
by a scaling factor 
smagnitude to vary its degree of contribution to the
template's overall response.
 is used to scale the magnitude of each edge
by a scaling factor 
smagnitude to vary its degree of contribution to the
template's overall response.
 
 
The value of v1 determines the degree of attenuation for misaligned edges. A higher v1 completely attenuates misaligned edges thereby approximating Sela's method of discarding the edges. However, when the Sobel operator is used to compute edge maps, the phase values are not reliable enough to have such a severe attenuation. Thus, a value of v1=2 was selected in our implementation.
 
 
 
 
 
 
