Abstract
We present a method for 2D shape matching usinga combination of distance functions and discrete curvature. Theeccentricity transform computes the longest geodesic distanceacross the object. This transform is invariant to translationand rotation. The maximal eccentricity points yield diametersacross the image. We compute the Euclidean distances from theboundary to the diameter to characterize the curvature of theshape. Our shape descriptor is comprised of the best matchesretrieved from the normalized histogram of the eccentricities, theHausdorff distance between the set of distances to the diameterand a measure of the number of points lying on either side of thediameter along with the peak values. We evaluate this descriptoron 2D image databases consisting of rigid and articulated shapesby ranking the number of matches. In almost all cases, the shapesare matched with at least one shape from the same class.
Reference
Ramachandran, G., & Kropatsch, W. (2015). A combined distance measure for 2D shape matching. In International Conference on Computer Vision and Image Analysis Applications. International Conference on Computer Vision and Image Analysis, ICCVIA2015, Tunesien, Suisse, Non-EU. IEEE: Computer Vision and Image Analysis Applications (ICCVIA), 2015 International Conference on. https://doi.org/10.1109/iccvia.2015.7351875