Our theoretical analysis indicates that our connected components
algorithm is scalable whenever 
, where c is
approximately 26 from the first expression in (11).
We have implemented our algorithm in SPLIT-C on the CM-5; the
resulting performance is plotted for images ranging from 
to 
pixels in size in
Figures 15 - 17 for p=16, 32, and
64 processors. Figure 3 presents the summary on
the performance of our connected components algorithm on the CM-5 and
clearly shows the scalability of our algorithm. Comparable results
for execution on the IBM SP-1 are given in
Figure 19, and for the IBM SP-2 in
Figure 21.