Sample-Based Synthesis of Illustrative Patterns

We present an improved method for synthesis of patterns defined as 2D collection of vector elements. Current solutions to this problem rely on triangulation of the input space or statistical measures of the sample to drive the synthesis step. We propose a method applicable to colored textures, from regular to stochastic, and which provides control over local density of elements. Also, our results show the same visual quality as previous works. The sample is segmented into groups of similar elements and we use a novel local neighborhood distance metric to compare distinct and incomplete neighborhoods. This metric does not ignores existing unpaired elements. The main synthesis loop consists of a procedural growth, where seeds are replaced by a reference to an element from the sample, generating new seeds until the target space is filled.