Automatic Digitizing of the Colour-Layer of Thematic Maps

A method for automatic classification and dcscription of the colour-layer of thcrnatic maps is presented. Owing to the type of noisc appcaring in the scanncd imagcs, we find it useful to combine pixel classificatiorl and rastcr-to-vector convcrsion in ordcr to produce acceptable results. l'hc colour value of an arca is dctcrmincd from a sclcctivcly sampled set of pixcls and a set of threshold valucs. I'ossihlc bordcrlines arc produccd by vcctorizing thc linc-network of the i m a g , and falsc bordcrlincs are rcmovcd by comparing ncighbouring sub-arcas. thcrcforc usually split into a numbcr of smaller areas by the linc network. Our proccdurc combines the two methods of pixcl classification and raster-to-vector conversion into an algorithm which is robust to most of thc noise and distortions commonly found. The proccdure has onc limitation: it rcquircs that thc borders of the colourcd arcas of intcrcst arc defined by lines, clearly distingrlishahlc from thc areas thcmsclvcs. THE COMBINED APPROACH Itxpcritncntal rcsults are discussed, and somc tcst maps arc prcA summary of the processing proccdurc which has becn uscd is ~cntr t l to illustr:ctc thc pcrformancc of the system. prcscntcd bclow. The raster-to-vector conversion system uscd is dcscribcd in morc dctail by Espclid and Eilcng in I I