Visually guided bottom-up table detection and segmentation in web documents

In the AllRight project, we are developing an algorithm for unsupervised table detection and segmentation that uses the visual rendition of a Web page rather than the HTML code. Our algorithm works bottom-up by grouping word bounding boxes into larger groups and uses a set of heuristics. It has already been implemented and a preliminary evaluation on about 6000 Web documents has been carried out.