HQ feature: descritores de forma utilizando a curva de Hilbert

Abstract

The similarity classification and retrieval based on shape descriptors are processes of great importance in pattern recognition and content-based shape retrieval tasks. The development of efficient shape descriptors is still a challenge. This work focuses on the development of new extractors to characterize shapes in 2D and 3D space using Hilbert space-filling curves and discrete wavelet transform decomposition. Several experiments are conducted using 2D and 3D silhouettes databases. The descriptors proposed are invariant under translation transformations and some experiments to assess the robustness on scale and rotation transformations and on occlusion, articulation and deformation effects were performed and obtained excellent results.