A new Rényi entropy-based local image descriptor for object recognition
Salvador Gabarda, Gabriel Cristóbal, P. Rodríguez, Carlos Miravet, Juan Manuel Del Cura · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2010
This paper shows how local directional entropy can be used as a tool to build up a robust local image descriptor for image feature extraction. Among other possible choices, the Rényi entropy has been selected as the main technique for this application. Local directional entropy which is related with the anisotropy images has been considered here as the basis for the design of a new Rényi entropy-based local image descriptor (RELID). The properties of this new descriptor are described and evaluated. The experimental results confirm that the new descriptor is endowed by most of the invariant properties desired for object recognition applications.