Benchmarking page segmentation algorithms

A method for automatically evaluating the quality of document page segmentation algorithms is introduced. Many different zoning techniques are now available, but there exists no robust method to benchmark and evaluate them reliably. Our proposed strategy is a region-based approach, in which segmentation results are compared with manually generated "ground truth files", describing all possible correct segmentations. A segmentation ground truthing scheme was already proposed. The evaluation of segmentation quality is achieved by testing the overlap between the two sets of regions. In fact, the regions are defined as being the valued pixels contained in the extracted polygons. An explicit specification of segmentation errors and a numerical evaluation are derived. The algorithm is simple and fast, and provides a multi-level output for each segmentation.