One of the complex documents in the real world is city maps. In these kinds of maps, text labels overlap by graphics with having a variety of fonts and styles in different orientations. Usually, text and graphic colour is not predefined due to various map publishers. In
More
One of the complex documents in the real world is city maps. In these kinds of maps, text labels overlap by graphics with having a variety of fonts and styles in different orientations. Usually, text and graphic colour is not predefined due to various map publishers. In most city maps, text and graphic lines form a single connected component. Moreover, the common regions of text and graphic lines have similar features; hence, the separation of text and graphic lines is a challenging task in document analysis. Generally, these text labels could not be recognized efficiently by current commercial OCR systems in city map processing. In this paper, we propose an image decomposition approach based on stroke width feature to extract text labels from city maps. In our approach, we assign to each pixel of image a local stroke width based on minimum distance from borders in four directional borders. This mapping generates a suitable representation to distinguish text and non-text pixels. The experimental results on several varieties of city maps are promising
Manuscript profile
