In order to extract medial axes of tree-like objects from three-dimensional images. A medial axis extraction algorithm is presented

it estimates coarse medial axis using centroids and applies multi-scale refinement to its main points

especially at bifurcations. The algorithm has both simplicity (as of centroid algorithms) and accuracy (as of scale space analysis methods) properties

and is suited in blurred and noisy conditions. The algorithm is successfully applied to the visualization of three-dimensional images of human heart that are acquired from medical CT. The method of generating curved section from medial axis and spreading it to screen is also given. Curved sectional display of branched vessels is firstly realized.