Shape Skeleton Classification Using Graph and Multi-scale Fractal Dimension

摘要

This paper presents a novel approach to shape characterization, where a shape skeleton is modelled as a dynamic graph, and its complexity is evaluated in a dynamic evolution context. Descriptors achieved by using this approach show to be efficient in the characterization of different shape patterns with different variations in their structure (such as, occlusion, articulation and missing parts). Experiments using a generic set of shapes are presented as also a comparison with traditional shape analysis methods, such as Fourier descriptors, Curvature, Zernike moments and Bouligand-Minkowski. Although the reduced amount of information present in the shape skeleton, results show that the method is efficient for shape characterization tasks, overcoming the traditional approaches.