Image-based Chinese seal carving generation with nonparametric sampling

This work presents a system for automatically generating Chinese seal carving works of desired style. By learning a training set of seal examples of the same style, the system can generate new seal carvings similar to this style through non-parametric sampling. Using K-nearest neighbor algorithm, stroke branches are interpolated to produce output strokes by linear combination. Each stroke branch is rendered by dragging a circular brush mask along the sweep of the Bspline which is presented with nonuniformly spaced interpolated points. We introduced a new sampling method, non-uniform sampling with constraint through hierarchical subdivision, to get interpolated points. The algorithm is simple and comprehensible in geometry. Experimental results show that this approach can obtain satisfying new articles of the desired styles.