Figure 9.35 (left and right) shows the performance of the synthetic system for the small noise protocol using polyline (see section 9.5) and shortest distance methods. Figure 9.35 (left) compares the mean error curves of the MRF vs. FCM (with smoother) using the PDM, while Fig. 9.35 (right) compares the mean error curves of the MRF vs. FCM (with smoother) using the shortest distance method. Using PDM, as the variance of the noise (a2) increases from 0 to 100, the mean error in both methods increases gradually. The mean error for the FCM (with smoother) remains under 1.6 pixels, while the mean error for MRF ranges between 1.6 and 1.8 pixels. The same pattern is observed using SDM method (see Fig. 9.35, right). It is also seen in the two graphs that FCM using PDM has a lower error compared to FCM using SDM.
In another procotol, we run the same PDM and SDM for FCM methods but with and without the Perona-Malik smoothing process. This can be seen in Fig. 9.36 (left and right). The method of PDE-based smoothing system improves
the error over non-PDE based system at large noise and thus is more robust in identification and detection process. It is also seen in the two graphs that FCM (with and without smoother) using PDM has a lower error compared to FCM (with and without smoother) using SDM.
In another procotol, we compare the MRF vs. FCM (without PDE smoother) and this can be seen in Fig. 9.37 (left and right). Using PDM, as the variance of the noise (a2) increases from 0 to 100, the mean error in both methods increases gradually. The mean error for the FCM (without smoother) remains under 1.6 pixels, while the mean error for MRF ranges between 1.6 and 1.8 pixels. The same pattern is observed using SDM method (see Fig. 9.37, right). It is also seen
in the two graphs that FCM (without smoother) and MRF using PDM have a lower error compared to FCM (without smoother) and MRF using SDM.
Was this article helpful?
Post a comment