Image Vectorization with Real-Time Thin-Plate Spline

Vector graphics becomes popular due to its compactness and scalability. However, current representations generally fall short in real-time editability and require intensive labor to create. To solve these problems, this work patchizes detailed features for localized and parallelized Thin-Plate Spline (TPS) interpolation in order to maintain interactive editing and manipulation abilities and to preserve complex image details. A user provides a patch labelling map to create explicit parametric patches for editability, and our system extracts an optimal set of curvilinear and transition features using the image’s intensity gradient distribution histogram for compact-ness and scalability. We optimally cluster extracted features into parameterized patches to parallelize TPS interpolation for real-time TPS kernel construction, inversion, and rasterization. Our real-time vector representation enables us to build an interactive system for detail-maintained image magnification, shape and color editing, abstraction and stylization, and material replacement of cross mapping. Experiments show that our proposed system can provide real-time editability while preserving structural and textural information better than raster-space operations and produce high-quality results comparable to existing state-of-the-art vector-based representations.