An Approach for Evaluating the Success of Information Extraction Procedures on Area Features

要旨

The present study proposes a conceptually simple yet effective approach for evaluating the success of information extraction procedures on area features, using five parameters: spatial error of omission, spatial error of commission, correctness, completeness and quality. To demonstrate this approach, three region based segmentation algorithms were applied to semiautomatically extract the outlines and areas of buildings from high-resolution satellite imagery. Qualitative and quantitative comparisons of the segmentation results demonstrate that the proposed parameters are able to evaluate the success of information extraction processes on area features. The five proposed parameters, supplemented with a shape index can be useful for evaluating result of information extraction processes on area features in all fields dealing with extraction of natural and manmade features.