Brush-Style Transfer through Constrained Convolutional Sparse Coding

Abstract

In this paper, a new technique to transfer the sketching style from one free-hand sketch to another is presented. Those sketches can be from different artists who used different brush styles while sketching. Given a unique brush-style from a sketch, our method transfers that style into another input sketch. Our brush-style transfer starts with an automatically constructed brush dictionary which proposes which sparse set of brushes are used at each part of the input sketch. After that, a oneto- one mapping is learned, between the unique brush sketch elements and the input sketch elements, by sparsely encoding input sketch with the brush dictionary. The quality of our brushstyle transfer is evaluated qualitatively by applying our technique on transferring multiple brush styles into input sketches. This is an addition to a quantitative evaluation through extensive subject case study were participants with 86% accuracy match a transferred sketch to the sketch they believe was used in the transfer.