A Two-Stage Framework for Polygon Retrieval

Abstract

We propose a two-stage framework for polygon retrieval which incorporates both qualitative and quantitative measures of polygons in the first and second stage respectively. The first stage uses Binary Shape Descriptor as a mean to prune the search space. The second stage uses any available polygon matching and similarity measuring technique to compare model polygons with the target polygon. This two-stage framework uses a combination of model-driven approach and data-driven approach. It is more efficient than model-driven approach since it reduces the number of polygons needed to be compared. By using binary string as index, it also avoids the difficulty and inefficiency of manipulating complex multi-dimensional index structure. This two-stage framework can be incorporated into image database systems for providing query-by-shape facility. We also propose two similarity measures for polygons, namely Multi-Resolution Area Matching and Minimum Circular Error Bound, which can be used in the second stage of the two-stage framework. We compare these two techniques with the Hausdorff Distance method and the Normalized Coordinate System method. Our experiments show that Multi-Resolution Area Matching technique is more efficient than the two methods and Minimum Circular Error Bound technique produces better polygon similarity measure than the two methods.