Modified Network Behaviour

The modifications I made to the resistive fuse network aid considerably the performance when this input image is moving. When a vertical edge segment is tripped, it also pre-biases edge segments to the left and right to expect a vertical edge. If this edge then moves horizontally, the edges will have a lower threshold than they normally would, and the edge is more likely to segment. This is shown in Figure 8.2.

Figure 8.2: Response of New Resistive Fuse Grid to Motion.
\begin{figure}\par
\centerline{\psfig{figure=fig/cool_motion.eps,width=5cm,angle=-90}}
\par\end{figure}

The old edge is still ``sticky'', and the new edge will segment before the old one starts to conduct. So for a time there will be two parallel edges. At this point, there will be a narrow line that is segmented on both sides. Unlike the narrow lines that are efficiently eliminated, this narrow line has gradients pointing in the same direction on both sides. However, the underlying voltage sources supplying these output pixels are dark. With the region segmented on both sides, there is nothing to prevent them being quickly pulled down to a low output value. This eliminates the voltage difference over the original edge, and so the edge starts conducting again.

So the time during which both edges have segmented is very small, and would appear almost instantaneous if viewed in real time. Furthermore, the edge will move only one pixel at a time, making the motion much easier to detect by further stages. So the modified network can potentially achieve a huge performance improvement over the simple resistive fuse network for real-time applications.

Matthew Exon 2004-05-23