SIMILAR-SHAPE RETRIEVAL IN SHAPE DATA MANAGEMENT

Retrieval of similar-looking shapes is an important problem in shape or visual database management. The authors describe the fundamentals of similar-shape retrieval and discuss the central issues that need to be resolved in designing a shape retrieval technique. They also present their own technique for retrieval of similar shapes, Feature index-Based Similar-Shape Retrieval (FIBSSR). This sys tem can handle images of articulated or rigid objects and query images of partially visible, overlapping, or touching objects. Query images can be simple or complex. In describing FIBSSR, the authors focus on shape representation, index structure, and query processing. Shape representation involves shape boundary and boundary points, called interest points. The shape boundary is coded as an ordered sequence of interest points. For the index structure, encoded feature vectors representing the shape boundary features help form a feature index for the shape database. The Euclidean distance between two feature vectors defines the similarity between the two respective features. And query processing involves query feature selection, formulation of possibly similar shapes, and formulation of the final response set. The authors also describe a prototype system based on FIBSSR to demonstrate how the key issues are resolved in the design of a typical shape retrieval system and to outline the important steps involved in its operation. A database of 101 rigid-and six articulated-shapes, with two components each, was used in several experiments to test this prototype system.