Abstract
This paper presents a new approach to similar shape retrieval by coarse-to-fine curve matching and data mining techniques. The proposed method extends the concepts of conventional data warehouse and image database for effective image indexing. An image data warehouse schema is developed to integrate multiple shape features for hierarchical shape representation, dynamic similarity measures and fast query processing. A guided search scheme is introduced, that combines the methods of invariant moments, 2D polygonal arc matching and B-spline curve matching, to search for the best similar shape in a hierarchical fashion. The proposed method is applied to the tropical cyclone pattern recognition by utilizing the active contour model (snake) to extend the traditional Dvorak technique for tropical cyclone intensity analysis and forecasting from satellite imagery.