Abstract

Image segmentation in RGB space is a notoriously difficult task where state-of-the-art methods are trained on thousands or even millions of annotated images. While the performance is impressive, it is still not perfect. We propose a novel image segmentation method, achieving similar segmentation quality but without training. Instead, we require an image sequence with a static camera and a single light source at varying positions, as used in for photometric stereo, for example. Here, foreground objects cast shadows onto background objects, the detection of transitions from light to shadow can be used to reveal the spatial structure of the scene and to trace the contour of an object. Inspired by interactive sketch colouring methods our novel image segmentation algorithm is based on Delaunay triangulations. After converting the pixel grid to a mesh, our algorithm operates on the face graph of the Delaunay triangulation where there is no notion of colour similarity. Instead, we rely on the edge length as a similarity indicator due to the circumcircle property of the Delaunay triangulation. Our method shows promising results without any training on annotated data.