ORGANIZING LARGE STRUCTURAL MODELBASES

We present a hierarchically structured approach to organizing large st ructural modelbases using an information theoretic criterion. Objects (patterns) are modeled in the form of random parametric structural des criptions (RPSDs), an extension of the parametric structural descripti on graph-theoretic formalism [1]. Objects in scenes are modeled as par ametric structural descriptions (PSDs). The organization process is dr iven by pairwise dissimilarity values between RPSDs. We also introduce the node pointer lists, which are computed offline during modelbase o rganization. During recognition, the only exponential matching process involved is between the scene PSD and the RPSD at the root of the org anized tree. Using the organized hierarchy along with the node pointer lists, the remaining work simplifies to a series of inexpensive linea r tests at the subsequent levels of the tree search. We develop the th eory and present three modelbases: 50 objects built from real image da ta, 100 CAD models, and 1000 synthetic abstract models.