Image retrieval by shape: a comparative study

Besides traditional applications (e.g. CAD/CAM and trademark registry), new multimedia applications, such as structured video, animation and the MPEG-7 standard, require the storage and management of well-defined objects. We focus on shape-based object retrieval and conduct a comparison study on four such techniques: Fourier descriptors (FD), the grid-based (GB) method, Delaunay triangulation (DT) and MBC-TPVAS (minimum bounding circles/touch-point vertex-angle sequence). Our results show that the similarity retrieval accuracy of our method (TVPAS) is as good as the other methods, while it has the lowest computational cost for generating the shape signatures of the objects. Moreover, it has low storage requirements and a comparable computation cost to compute the similarity between two shape signatures. In addition, TPVAS requires no normalization of the objects and is the only method that has direct support for RST (rotation/scaling/translation) query types. In this paper, we also introduce a new shape description taxonomy