Towards robust scene analysis: A versatile mid-level feature framework
David Joshua Engel, Cristóbal Curio · MPG.PuRe (Max Planck Society) · 2010
We present a novel set of shape-centered interest points. The interest points are formed at locations of high local symmetry. Our symmetry detection is based on Gradient Vector Flow (GVF) [1] fields which provide a high level of stability against noise. The shape centered interest points allow for a robust scale and orientation estimation. We have shown their usefulness for image encoding and superpixel segmentation and demonstrat that they carry information that is to a certain degree complementary to corner based interest points.