Improved Multi Angled Parallelism for separation of text from intersecting linear features in scanned topographic maps

Separation of the text and graphics layers in maps with dense and overlapping sets of features (e.g. topographic maps) is a challenging problem. Multi Angled Parallelism (MAP) provides an efficient tool to detect miscellaneous linear features using directional morphological operations and higher order feature representation. However, in its original formulation sides of characters, short lines, and parts of lines that pass through characters are often misclassified. This paper presents an improvement over MAP to automatically extract complete line networks with arbitrary orientation and curvature even when they pass through characters with minimal impact on the text content. The resulting text only image can then be processed for text grouping, reorientation, and recognition. The proposed algorithm does not rely on heuristics and can be easily adapted to work with maps of various scales and sources and other line drawing images by adjusting only a small number of parameters.