Large three-dimensional arrays of data are routinely generated from several investigative sources, e. g., the medical, meteorological, metallurgical, and geological sciences. The data values in the arrays are representative of some intrinsic property of the physical medium being investigated. Visualization of such "solid" data can be made easier by the extraction of surface boundary and volume information from the cube of sampled data: we want to be able to "see" interior objects bounded by detected surfaces. When surface boundaries are detected, data exterior to the objects should be ignored and three-dimensional images of the object displayed. Detecting surface boundaries and displaying them interactively gives scientists the ability to find trends and inflection points in large datasets. 

Most of the work done in surface boundary detection has been done on medical data. Medical datasets are relatively small and well behaved compared to data from other popular sources. Tree - dimensional migrated seismic data sets can be orders of magnitude larger. The data values are noisier and the histograms much less well defined. 

The algorithms and data structures described here allow users to select, collect and display surface boundaries found in tree-dimensional data sets of arbitrary size and value ranges. The surface boundary detection/volume collection algorithm is unique from previous approaches in several aspects. Its recursive investigation of voxel neighbor provides a way to seek out even the most complex surface boundary. The current program allows the user to select from several voxel "neighbor" groups and to use classification functions that make it possible to successfully investigate a wide variety of data. Previous methods work well with only certain types of volume data or rely on special hardware for surface detection or display.