Multiple-seed 3D connectivity filling for inaccurate borders

Abstract

Reconstructing a volume representation in the presence of inaccurate boundary data is becoming a fundamental issue in a variety of fields, such as modeling of natural objects, medical imaging, graphics, and robotics. Given boundary data, parts of which are incorrect, a low-level representation of the enclosed volume in terms of octrees is found. This representation is useful in providing accurate volume estimates related to the enclosing boundary. The connectivity filling technique proposed here is, basically, a region-growing technique which simultaneously moves from the border elements toward the outer and inner normals: in the first case it propagates the label white and in the second case the label black. A labeling component algorithm and a border evaluation technique are then used to resolve ambiguities that may arise when gaps or superpositions of border elements occur. An example has been especially designed to illustrate the problem and its proposed solution. Illustrations of the algorithm's capability applied to two natural objects conclude the presentation.