Junction Detection and Multi-orientation Analysis Using Streamlines

We present a novel method to detect multimodal regions composed of linear structures and measure the orientations in these regions, i.e. at line X-sings, T-junctions and Y-forks. In such complex regions an orientation detector should unmix the contributions of the unimodal structures. In our approach we first define a (streamline) divergence metric and apply it to our streamline field to detect junctions. After this step we select all streamlines that intersect a circle of radius r around the junction twice, cluster the intersection points and compute the direction per cluster. This yields a multimodal descriptor of the local orientations in the vicinity of the detected junctions. The method is suited for global analysis and has moderate memory requirements.