Multi-scale Vessel Boundary Detection
International Conference on Computer Vision Workshops: Computer Vision for Biomedical Image Applications (ICCVW)
In this paper, we present a robust and accurate method for the segmentation of cross-sectional boundaries of vessels found in contrast-enhanced images. The proposed algorithm first detects the edges along 1D rays in multiple scales by using mean-shift analysis. Second, edges from different scales are accurately and efficiently combined by using the properties of mean-shift clustering. Third, boundaries of vessel cross-sections are obtained by using local and global perceptual edge grouping and elliptical shape verification. The proposed algorithm is stable to (i) the case where the vessel is surrounded by other vessels or other high contrast structures, (iii) contrast variations in vessel boundary, and (iii) variations in the vessel size and shape. The accuracy of the algorithm is shown on several examples.