| A syntactic/semantic curve matching method is developed at two levels of similarity and affine invariance. The developed method is examined for matching of smooth, almost planar, almost rigid curves. Based on the introduced curve matching method, a computer-assisted system for photo-identification of several species of marine mammals is designed. The identification performance of the system is examined on databases of bottlenose dolphins' dorsal fin images and sperm whales' fluke images, and compared to that of curvature matching as well as a few other existing matching methods.; The syntactic/semantic curve matching system consists of a string representation scheme, and a string matching strategy. An attributed low-level string representation of a curve is constructed from its curvature versus arc-length signature. A high-level string representation is built over the low-level one via merging appropriate groupings of primitives, emphasizing on significant structures on the curve. A family of syntactic/semantic distance measures between two strings is introduced and used as a dissimilarity measure for database search, highlighting both the syntactic and semantic differences.; The developed method is proved to be effective for smooth planar curves, which does not possess many sharp distinguished points. First hit matching rates of 55% and 60% on databases of dolphin and whale images are achieved using the similarity and affine invariant syntactic/semantic curve matching methods, respectively. Both statistics surpass curvature matching and other examined methods. It is shown that the identification performance is not sensitive to perturbations of the design parameters around their nominal values. Comparing to curvature matching, the computation speed is improved by approximately an order of magnitude. For those curves that are incomplete at the end, the developed method is proved to perform as good as curvature matching. However, when a significant structure is hidden, curvature matching appears to perform better.; The photo-identification results for dolphins and whales indicate that the developed matching method outperforms all of the previous identification methods. The developed computer-assisted photo-identification system is found to help marine mammalogists identifying an individual animal, since it allows them to examine only a handful of candidate images instead of the manual searching of the entire database. |
