Abstract: A complete contour-based reconstruction method must establish correspondence, solve branching problems,and construct tiles. Most modern reconstruction algorithms typically address only one or two of these problems. Therefore, their applications do not achieve complete solutions with complicated objects, such as the considerably convoluted and highly branched cortex of the human brain extracted from Magnetic Resonance Imaging(MRI) data. This paper presented an efficient Voxel-coding algorithm; which can handle complicated many-to-many branching and holes in a fully automatic and systematic way. First the contours from adjacent slices are projected onto an intermediate plane. And then be divided into groups on the basis of their difference regions. For each group of contours, skeletons arc extracted from the corresponding region. These skeletons are used to measure contour dissimilarity and to decompose dissimilar or complicated branching contours into simple and similar contour-to-skeleton pairs. Reconstructed surfaces arc 2D manifold triangle meshes which pass only input contours along slices. The algorithm has been tested using both hand-made data and real complex human cortical MRI data, demonstrating its efficiency.

Key words: Contour, Surface reconstruction, Branching, Skeleton, Voxel-coding, Contour skeleton match