A rapid 2-D centerline extraction method based on tensor voting

Centerline extraction is widely used in medical image processing. It can benefit applications such as building the connectivity map of neurons from microscopic images as well as examining retina vessels for preventing blindness. Many methods have been developed to extract centerlines from 2-D images. An algorithm based on 2-D rapid tensor voting is proposed in this paper. This method uses the Canny edge detector and a simple ridge finding algorithm to roughly extract centerlines, which is fast, does not require any seeds and allows the object to be disconnected. Then efficient 2-D tensor voting is applied to enhance the centerline, which can rapidly bridge the gaps caused by the earlier step and reject artifacts due to noise. We demonstrate the robustness of the algorithm and compare with existing methods. The result shows good computational efficiency as well as accuracy.