A picture-descriptor extraction program using ship silhouettes

Abstract

This research examines the practicality of automatically identifying the features of the major structures of ship silhouettes using rule-based extraction and identification techniques. The process was broken into three phases: (a) finding the silhouette boundary, (b) locating the "bumps" (apparent superstructures) on the boundary, and (c) describing the bump features qualitatively using a multidimensional-feature-space classification. The program for the first phase is written in C while the programs in the other two phases are written in MPROLOG and run on a Motorola 68020-based workstation. The programs accurately identified 78% of all bumps examined on six ships. The lists of bump descriptions showed the key differences between two different ships of the same ship-class, indicating future programs could identify ships using the output from the programs of this thesis.