Extracting Words and Multi-Part Symbols in Graphics Rich Documents

摘要: We present an algorithm for grouping multipart symbols, dashed lines, and character strings extraction from line drawings. The image undergoes a lossless raster-to-vector conversion creating as its vector representation undirected graph, so-called run graph. Next, the elements of graph are extracted classified probabilistically based upon their geometric features using decision tree. An area Voronoi tessellation members sets is constructed, which neighborhood derived, guaranteed to be minimal complete. then traversed group various input different recognition modules. No priori font or other domain specific information required grouping, no special geometrical relationships among assumed. Results presented with example images taken those used by our Swiss cadastral map understanding system.