The two techniques have several parameters. The values of
these parameters considerably affect the performance of the
algorithms. So, for each algorithm, we varied the values of the
parameter to be tuned while fixing the values of the other
parameters (initially these values were selected in the
literature). The numerical results obtained using both
algorithms PSO and DE with c=2, 3 and 4 are presented in
Table II and III. The optimal thresholds found using the two
algorithms are: T, n, T2 and T3. The performances of both
approaches were compared using the fidelity criterion: the
peak-to-signal-noise (PSNR) ratio. As it can be seen, when the
number of thresholds is small or great, both algorithms give
good results in terms of accuracy and robustness. However, we
can see that the PSO is the most efficient in terms of time
execution. Statistical results in table II and III suggest that
PSNR values of the best solutions are obtained from PSO
algorithm and the best time processing is providing by PSO
algorithm too (See Fig 2).
The optimal thresholds obtained by the proposed algorithm
were compared with those found in the work [9]. Both PSO
and DE algorithms provide the same optimal values of
thresholds, but the processing time is reduced for the proposed
approaches especially for multi-level Thresholding. As it can
be seen, the Fuzzy c-partition entropy using PSO and DE
algorithms perform equally well in terms of the quality of
image segmentation (Fig 3) and leads to a good visual result
with processing time reduced.