Automatic Feature Extraction and Text Recognition From Scanned Topographic Maps

A system for automatic extraction of various feature layers and recognition of the text content of scanned topographic maps is presented here. Linear features which are often intersecting with the text are first extracted using a novel line representation method and a set of directional morphological operations. Other graphical objects are then removed in several stages to obtain a text-only image. A custom defect model is subsequently used to create an artificial training set for a Hidden Markov Model-based character recognition engine. Finally, the recovered text is recognized using this multifont segmentation-free optical character recognition (OCR). Extensive testing is conducted to assess the performance of different stages of the proposed system. Furthermore, our custom OCR is shown to achieve a 94% recognition rate for the extracted text, thereby outperforming a commercial OCR used as a benchmark.