Capturing Raster Data From Scanned Thematic Maps Using Desktop Graphics Software

Abstract

Data capture from paper maps remain a slow and costly part of many Geographic Information System projects. In this paper a simple method is described which makes use of the facilities of desktop graphics software to capture raster data from scanned thematic maps. The first stage is to use the flood fill tool to identify the regions of the map that represent the data values and set them to a separate colour for each thematic category. This edited image is then imported into a GIS, where a specially written majority filter is used to classify any remaining pixels to produce a raster GIS layer. The technique is tested on samples from three maps, giving accuracy levels of over 90% in all cases when compared with digitising the same maps by hand. The technique could be useful for situations where a rapid and simple means of data capture is needed and could be developed further by drawing on techniques for contextual classification rather than relying on a general purpose flood fill algorithm.

Forest cover mask from historical topographic maps based on image processing

“…For tested map samples, HSV colour space provided better results (by 1–2% of correctly recognized pixels) than CIELab space used by Herrault et al . () or RGB space that was employed by Wise ().…”

“…These results are similar or better than the accuracies of feature extraction received in other studies. For instance Wise (, ) obtained accuracy levels of 80–90% compared with digitising the same maps by hand and Herrault et al . () received similar Kappa values in their study.…”

“…Our methods focused on forest areas that are shown on maps with distinct colour. We received accuracies similar to results of other studies which aimed to extract two‐dimensional features from scanned paper maps (Wise, , ; Herrault et al ., ), proving the effectiveness of our approach. In cases, where graphical, structural or textural forest representation is to be taken into account, potential of other methods might be higher (Leyk et al ., ; Leyk and Boesch, , ).…”

“…contour lines that were later processed into Digital Elevation Models, rivers or roads (Kaneko, ; Arrighi and Soille, ; Wu et al ., ; Ghircoias and Brad, ; Oka et al ., ). For two‐dimensional objects (polygons) the automated approaches are more complex (Wise, , ; Leyk and Boesch, ). Two‐dimensional objects are typically obscured by various other map symbols.…”