This paper presents a method for the extraction of land-use classes from colour thematic maps. The material used here are 1:50000 thematic maps of IBGE (Instituto Brasileiro de Geografia e Estatistíca). Land cover classes within this map are not only characterized by a different colour, but also by repetition of coloured patterns. And, as the coloured areas concerns, they are not solid, but more a combination of solid colours and speckling, which results in a certain texture.
The first step was to scan the map and produce a raster matrix. For this purpose, a low cost flat-bed optical scanner was used. The image was segmented according to its colours using a simple approach. Pixels representing different land-use classes were isolated and binary images were obtained.
An approach based on the Delaunay triangulation was used to correct errors and extract land-use areas. The labeled pixels were grouped into homogeneous regions according to a measure of proximity. Finally, mathematical morphology was used to improve the quality of the borders of the polygons. At the end, different land-use classes were extracted from the map based on their colour and texture.
The paper includes an example of the application of this methodology. A 512 x 512 image at different steps is included. The resulting extraction and digitizing procedure is still not as good as a hand-made one, especially when there is a need for high accuracy, however its advantages are speed and automation.