A framework of perceptual features for the characterisation of 3D textured images

This paper presents a multiresolution system for volumetric texture analysis. The originality of this system partially originates from its use of combinations of perceptual texture features that correspond to adjectives commonly used by humans to describe textures. To approximate these features, we use a combination of different families of texture analysis methods rather than a single texture analysis model. This choice is necessary to obtain a good perceptual feature approximation and allows our system to be robust and generic. Moreover, by using our human-understandable features (HUF), it is convenient for a user to manipulate and select the features that are, according to the user, relevant for a given application. Two experiments are presented: the first experiment demonstrates the strong correspondence between our features and a human’s description of textures, and the second demonstrates the performance of our proposed method. Finally, the proLudovic Paulhac Laboratoire Informatique de l’Université François Rabelais de Tours 64 avenue Jean Portalis 37200 Tours Tel.: +332 47 36 14 14 Fax: +332 47 36 14 22 E-mail: [email protected] Pascal Makris Laboratoire Informatique de l’Université François Rabelais de Tours E-mail: [email protected] Jean-Yves Ramel Laboratoire Informatique de l’Université François Rabelais de Tours E-mail: [email protected] Jean-Marc Gregoire UMR INSERM U930, CNRS ERL 3106, équipe 5, Université François Rabelais de Tours E-mail: [email protected] posed HUF are integrated into an interactive segmentation system and are compared to previously proposed descriptors through analysis of several segmentation results of 3D ultrasound images.