Image Processing Reference
In-Depth Information
this case, unless all the pixels in the kth layer are checked, the deletion of
pixels in a deeper layer (i.e., (k + 1)th layer) cannot start. Further, to ensure
the quality of the thinned pattern, medial points act as anchor points that are
never deleted. They are connected at the end of the computation. In [204],
these anchor points are taken from the RCMD of the object. The computation
is described in Algorithm 12.
m
4
m
3
m
2
m
5
m
1
q
m
6
m
7
m
8
FIGURE 6.10: Neighborhood variables of a point q.
Algorithm 12: Skeletonization from MAT of a 2-D Object
Algorithm: Skeletonization from MAT (SMAT)
Input: The Object Σ, The distance Transform DT, The MAT M. Let
D
max
be the maximum distance value in DT.
Output: Skeleton S.
1. Obtain the RCMD of the object. Let us denote the set as Γ.
2. For k = 1 to D
max
{
2a. For each pixel p,
i. If (DT(p) = k) and p has a 4-neighbor in the background,
compute X
H
(p).
ii. If (X
H
(p) = 1)||(p ∈ Γ) retain the point and include it in S,
else delete it.
2b. Perform step 2a till all the points at level k are checked.
}
End Skeletonization from MAT (SMAT)
It was observed in [204], that there could be at most two scans required for
each level k in Step 2 of the above algorithm. The other advantage of having a
distance transform in this case is that it is convenient to perform morphological
