Abstract
Abstract
Improvements are described for a pattern recognition system for recognizing shipbuilding parts. This is achieved by using a new simple and accurate corner finder. The new system initially finds corners in an edge detected image of a ship's part and uses that new information to extract Fourier descriptors to feed into a neural network to make decisions about shapes. Results show that the new corner finder was better at distinguishing between various ships' parts than other corner finders and proved to be a valid approach. The new corner finding technique uses a bottom-up approach to find corners by sampling points in edge-detected images and calculating the distance between the endpoints of a window around each sampled point. The points with the minimum distance are then interpreted as corners. Using an all-or-nothing accuracy measure, the new corner finding technique achieved an improvement over other systems. The new corner finder was included as pre-processing before extracting Fourier descriptors and using the artificial neural networks to identify parts. The whole system recognized parts more quickly and more efficiently than the most recently published systems.