Image Vectorization and Shape Analysis

Image vectorization converts raster images into parametric representations that faithfully encode their prominent geometric features. Unlike pixel-based formats, vector representations are resolution-independent, scale-invariant, and amenable to geometric analysis. We develop variational methods that produce interpretable and compact parametric descriptions of image content, with applications in digital illustration, geometric analysis of discrete objects, and scalable rendering. —

Shape Vectorization by Affine Shortening Flow

We leverage affine shortening flow to remove pixelization artifacts while preserving the scale-invariant geometric features of image contours.

Compared to commercial software, our approach yields more interpretable and efficienct vector graphics.

References

(He et al., 2023) (He et al., 2022) (He et al., 2021)

Topological-aware Color Image Vectorization

By careful analysis of local topological patterns in raster images, we develop a surgical strategy for image vectorization that preserves prominent singularities with high fidelity.

Our approach produces vector graphics with high fidelity.

References

(He et al., 2023)