With the growth in computing, storage and networking infrastructure, it is becoming increasingly feasible for multimedia professionals – such as graphic designers in commercial, manufacturing, scientific and entertainment areas - to work with 3D digital models of the objects with which they deal in their domain. Unfortunately most of these models exist in individual repositories, and are not accessible to geographically distributed professionals who are in need of them. 

Building an efficient digital library system presents a number of challenges.  In particular, the following issues need to be addressed

·         What is the best way of representing 3D models in a digital library, so that the searches can be done faster?

·         How to compress and deliver the 3D models to reduce the storage and bandwidth requirements?

·         How can we represent the user’s view on similarity between two objects?

·         What search types can be used to enhance the usability of the digital library and how can we implement these searches, what are the trade-offs?

In this research, we have developed a digital library architecture for 3D models that addresses the above issues as well as other technical issues.  We have developed a prototype for our 3D digital library (3DLIB) that supports compressed storage, along with retrieval of 3D models. The prototype also supports search and discovery services that are targeted for 3-D models. The key to 3DLIB is a representation of a 3D model that is based on "surface signatures". This representation captures the shape information of any free-form surface and encodes it into a set of 2D images. We have developed a shape similarity search technique that uses the signature images to compare 3D models. One advantage of the proposed technique is that it works in the compressed domain, thus it eliminates the need for uncompressing in content-based search. Moreover, we have developed an efficient discovery service consisting of a multi-level hierarchical browsing service that enables users to navigate large sets of 3D models. To implement this targeted browsing (find an object that is similar to a given object in a large collection through browsing) we abstract a large set of 3D models to a small set of representative models (key models). The abstraction is based on shape similarity and uses specially tailored clustering techniques.  The browsing service applies clustering recursively to limit the number of key models that a user views at any time.

We have evaluated the performance of our digital library services using the Princeton Shape Benchmark (PSB) and it shows significantly better precision and recall, as compared to other approaches.