The size and complexity of scientific and medical datasets have always presented a challenge to interactive 3D visualization systems. In general, high quality images in this setting require graphics hardware acceleration to achieve the rendering speeds required for reasonably low interaction latency. A further challenge arises as the World Wide Web emerges as a mechanism for broad dissemination of these datasets in educational and collaborative applications (see for example [2], [20], [1]). A Web-based interactive 3D visualization system that is targeted at average users faces two significant constraints: the transmission cost to the client must be low, and the user must be able achieve reasonable performance on an average desktop PC without specialized hardware. In most existing systems, however, transmission costs are proportional to the complexity of the underlying geometry, and rendering occurs on the client. This configuration is acceptable for interacting with simple 3D worlds, but impractical for large datasets.
Image-based techniques transmit multiple 2D images of the 3D scene for interpolation by the client, and have succeeded in lowering transmission costs and simplifying the computation that must be performed by the client during interaction. However most existing techniques, image-based as well as others, view interactive 3D visualization within the VR paradigm, in which navigation is the primary framework for interaction. For medical and CAD applications in particular, other types of interaction are of equal if not greater importance. These include the ability to turn structures off or adjust their transparency to reveal what lies beneath them, the option to interactively tag important structures with meaningful colors, and the capability to perform quick picking of individual structures.
The development of the multilayer image (MLI) was motivated by our desire for a Web-based system that allows the average user to interact with high quality 3D scenes derived from large datasets. Such datasets are typical in medical applications, where a high level of geometric complexity is essential for evaluation of patient data by radiologists and surgeons. Unfortunately, due to their size, these datasets are ill suited for transmission over low-bandwidth network connections and rendering on desktop PCs.
Our new image-based technique utilizes MLIs to support several important types of interaction, while imposing low transmission costs that are comparable to the cost of transmitting several still images, and requiring only minimal computation by the client for display and update. Our technique cannot support navigation or multi-resolution zoom capabilities directly; however we believe that for a large number of applications, the lack of these two types of interaction can be significantly offset by providing multiple well-selected fixed views in which the user can remove or make semi-transparent those structures that are inconveniently located in any particular view.