Face Sketch Synthesis with Joint Training Model
Weiguo Wan, Hyo Jong Lee · 2017
Face sketch synthesis plays an important role in both law enforcement and digital entertainment. Frequency-used exemplar-based methods mainly consist of neighbor selection and reconstruction weight representation two parts. However, the training sketches usually are not taken into account during both of the aforementioned two processes. Thus, the obtained reconstruction weight may not fit for the sketch synthesis due to the difference between the selected training photo and sketch patch pairs. In this paper, we propose a joint training model to improve the sketch synthesis performance by concatenating the high-pass information of the training sketch patches with photo patches together when calculating the reconstruction weight. Extensive experiments on public dataset show that the proposed model can obtain outperform target sketches both from subjective and objective comparison.