Hierarchical diffusion curves for accurate automatic image vectorization

Guofu Xie, Xin Sun, Xin Tong, Derek Nowrouzezahrai
2014 ACM Transactions on Graphics  
Links: DL PDF Previous approaches use edge detection to place curves along sharp image features, relying on the assumption that regions with high Laplacian magnitude will overlap intensity discontinuities in the image domain. We show that, although this assumption is valid for sharp edges, it does not hold in general. Specifically, for blurred edges or bumps, the image edges do not overlap the Laplacian maxima (e.g, Fig. 3 ). We observe this behavior in both natural images and computer
more » ... ed imagery resulting e.g. from a turbulent fluid simulation. We formulate image vectorization as a Laplacian (and/or bilaplacian) reconstruction problem, and seek to fit sparse and connected curves to directly approximate the image Laplacian/bilaplacian. We avoid traditional edge detection in favor of extracting curves in the
doi:10.1145/2661229.2661275 fatcat:lknq4cwpwbhvdo6k456nkg2thq