Shape characterization using laws of texture energy measures facilitating retrieval

Feature characterization schemes catering shape indexing and retrieval have been a subject undergoing intense study in computer vision. Here, a feature characterization scheme is presented using the Laws of Texture Energy Measures targeting shape retrieval. The LTEM-based descriptor refines edges of shape images to produce highly discriminative features. Later, a feature representation arrangement packs it into global-structural shape histograms that are, subsequently, used for matching and retrieval. Exhaustive experiments of the resulting descriptor across the MPEG-7, Tari-1000 and Kimia's 99 datasets render consistent Bull's Eye Retrieval rate of 90%, revealing its highly distinctive nature among the intra-and inter-shape classes. Moreover, the witnessed BER clearly indicates that the descriptor is robust to different affine transformations.