Detection of perceptual junctions by curve partitioning and grouping

This paper presents a perceptual organization based method for the representation and extraction of junction structures of edge segments from digital images. Perceptual Junctions (PJs) are higher-level view invariant feature entities, which are made up by intersected generic edge tokens including both linear and non-linear segments. The class of low-order PJs (LPJs) is the junctions defined by two connected segments, and detected directly by an edge tracking and partitioning algorithm. The class of high-order PJs (HPJs) is the junctions made up by more than two segments which are extended from LPJs by grouping additional segments from different edge traces. The method is robust since it mainly uses qualitative perceptual features. The computation is efficient because it is mainly involved in symbolic reasoning. The experimental results are provided.