Using a low-level representation of images, like matching pursuit, we introduce a new way of describing objects through a general description using a translation, rotation, and isotropic scale invariant dictionary of basis functions. We then use this description as a predefined dictionary of the object to conduct a shape recognition task. We show some promising results for the detection with simple shapes.
Martin Jaggi, Sebastian Urban Stich, Sai Praneeth Reddy Karimireddy, Anant Raj