Face Sketch Synthesis Based on Rough Drawing

Existing face sketch usually requires well-trained artist to draw natural and realistic images,which hinders its wide usages in digital entertainment and public security management.We present an approach to automatically generate face sketches based on face rough drawing.Firstly,the input face rough drawing is divided into many overlapped patches.Then,the relation of patches between the rough drawing and the sketch is obtained by a coupled dictionary,which is learned from a set of training data using locality-constrained linear coding.Finally,using simulated annealing algorithm,sketches are synthesized from selected candidate sketch patches and optimized by a global energy optimization based on a Markov random field framework.We evaluate our approach on the face sketch XM2 VTS database.Experimental results validate the effectiveness of the proposed method.