Clustering Techniques for Out-of-Core Multi-resolution Modeling

Abstract

Thanks to improvements in simulation tools, high resolution scanning facilities and multidimensional medical imaging, huge datasets are commonly available. Multi-resolution models manage the complexity of such data sets, by varying resolution and focusing detail in specific areas of interests. Since many currently available data sets cannot fit in main memory, the need arises to design data structures, construction and query algorithms for multi-resolution models which work in secondary memory. Several techniques have been proposed in the literature for outof-core simplification of triangle meshes, while much fewer techniques support multi-resolution modeling. Some such techniques only deal with terrain data [2, 8, 10, 11]. Techniques proposed in [3, 6, 7, 9, 14] have been developed for free-form surface modeling and most of them are based on space partitioning. Our goal is to design and develop a general technique for irregularly distribuited data describing two and three-dimension scalar fields and free-form surfaces. In the spirit of our previous work, we define a general out-of-core strategy for a model that is independent of both the dimension and the specific simplification strategy used to generate it, i.e., the Multi-Tessellation (MT) [12, 5]. The MT consists of a coarse mesh plus a collection of refinement modifications organized according to a dependency relation, which guides extracting topologically consistent meshes at variable resolution. We have shown that the other multi-resolution data structures developed in the literature are specific instances of an MT. Thus, data structures optimized on the basis of a specific simplification operator, like edge collapse or vertex removal, could be derived from a general out-of-core MT. The basic queries on a multi-resolution model are instances of selective refinement, which consists of extracting adaptive meshes of minimal size according to application-dependent requirements. We have first analyzed the I/O operations performed by selective refinement algorithms and designed and implemented a simulation environment which allows us to evaluate a large number of data structures for encoding a MT out-of-core. We have designed and developed more than sixty clustering techniques for the modifications forming a MT, which take into account their mutual dependency relations and their arrangement in space. Based on the data structure selected through this investigation, we are currently developing an out-of-core prototype system for multi-resolution modeling which is independent of the way single modifications are encoded.