Prototypicality effects in global semantic description of objects

We introduce a new global semantic descriptor for objects based on the prototypicality effects of the Prototype Theory. Our descriptor encodes and stores the semantic meaning of the object's category while describing its features using the semantic prototype computed with CNN-classifications models. Our method uses semantic prototypes to create discriminative descriptor signatures while simulating the prototypicality organization within a given category. A dimensionality reduction function has been created to keep the semantic meaning within the signature of our descriptor. Our experiments showed that i) our descriptor preserved the semantic information used by the CNN-models for classification tasks and ii) our distance metric enables the simulation of the prototypicality organization.