Database-Retrieval Oriented Approach for Model-Based Object Recognition

Model-based recognition of objects when the number of model objects becomes large is a challenging problem which makes it increasingly difficult to view the object recognition problem as - “find the best match” problem. We present a database-retrieval oriented approach for this case wherein the goal is to index, retrieve, rank and output a few top-ranked models, according to their similarity with an input query object. The approach consists of three stages: (1) feature-based representation of model objects and object-feature correspondence analysis; (2) clustering and indexing of the model objects in the factor space; and (3) ranking indexed models based on mutual information with query object. We believe that this approach has practical advantages for large model databases where the best-match approach can be unreliable. It also finds applications in semi-automatic object recognition tasks which involve interaction with humans. Experimental results are presented using a variety of data to demonstrate the merits of our approach.