Abstract:
In this paper we present an approach to indoor classification that presupposes a certain amount of prior information in terms of statistical information about possible interdependencies among objects and locations. A preselective perception process (that here is only hinted), using a database of textures, and built using the energy value computed with a tree-structured wavelet transform, selects regions in the image and, according to the database, builds an observation state including also a saliency map of the features of the image. The process of delivering this information is interleaved with a process that using a database of interdependencies between objects and locations, mimicking a memory, forms an hypothesis about the current location. We show that the process of hypothesis formation converges under specific constraints.