3d Model Retrieval Based On Optimal Search Theory

The structures of 3D shapes can be easily acquired by the advanced modeling and visualizing techniques. This leads to an exponentially growth of 3D models in the database. To overcome the disadvantages of text-based retrieval, researchers develop a number of Content Based Retrieval (CBR) systems. The main idea for content-based retrieval is to create a set of features that can efficiently describe 3D models. We often use high-dimension feature vectors to achieve this goal, however, the large amount computation will make exhaustive retrieval infeasible. Thus it is a big challenge to find an efficient way to search 3D models under a limited resource. The search resource usually refers to the time and money spent in the retrieval.To find an efficient strategy for 3D model retrieval task, we refer to optimal search theory. Optimal search theory is a theory of finding the "best" way to detect a pre-claimed object, usually called the "target". Pioneered by G. Kimball and B. Koopman in 1940's, optimal search theory is developed from the statistical decision theory in Operations Research. Now it has been widely used in military, industrial, agriculture, criminology, market investigation, census, medical research, and so on. We may categorize optimal search problems according to whether the space and time are discrete or continuous, and whether the target is stationary or mobile. In this thesis, we study the situation with stationary target in the discrete space.Firstly, we consider the 3D model retrieval problem under a limited time resource. We assume that the total search time was bounded and hence the retrieval task will not succeed for sure. In this case, we derive the optimal strategy with which the task could succeed with the highest detection probability. In particular, we give the method to compute the initial probability distribution of target's location and the form of detection function. Thus we formulate the original problem as an optimal search problem. Our experimental results show that the optimal search strategy is efficient for information retrieval on structured database. Secondly, we study the situation when false targets exist during retrieval. To distinguish false targets from the real one, a two-step search...