Raster-To-Vector Conversion: A Trend Line Intersection Approach to Junction Enhancement

One of the problems with automatic or semi-automatic raster-to-vector conversion is dimpled junctions, which are caused by the underlying thinning algorithms. While useful for shape reconstruction from skeleton, dimpled junctions are undesirable for data conversion. Techniques, referred to as junction enhancement, have been developed to deal with, among others, the dimpled junction problem. However, existing junction enhancementtechniques lack well-defined procedures and are only able to handle junctions on a case by case basis. This paper presents a trend line intersection approach to systematically enhance all types ofjunctions. This approach has a set of simple and well defined procedures for estimating the trend lines of all arcs coming to a junction, and for calculating the new junction location as the intersection of all the estimated trend lines. It can handle junctions with any number ofincoming arcs and any type ofconfiguration using the single set of procedures. The experiments have shown that the new approach is very effective in dealing with dimpledjunctions