3D shape retrieval in distributed databases

As the size of 3D object databases grows rapidly, the emerging issue related to 3D shape retrieval in a distributed database is interesting to consider. In this paper, we propose extended methods designed for centralized databases, especially those based upon BagofFeatures (BoF) approach for shape indexing. Such methods use a visual dictionary (codebook) for computing the descriptor of a given 3D object. This codebook is obtained by clustering a set of features obtained from all the objects of the database. Such clustering, however, does not suit distributed databases very well, particularly because it gets too timeconsuming. Our solution consists in associating a local codebook to each database and then generating a distributed codebook from the local codebooks only instead of directly dealing with all the objects of each distributed database. Experimental results show that the obtained relevance with our proposed approach is very close to the case of a centralized codebook.