Towards robust face sketch synthesis with style transfer algorithms

Abstract

We propose an approach for face sketch synthesis by employing deep image transformations using an artistic style transfer algorithm. Face sketch synthesis remains an area of great interest in the research community as well as its applications in law enforcement towards face recognition. Recent methods for this problem typically employ traditional approaches to synthesize face sketches to digital images. However, most approaches are gradually shifting towards convolutional neural networks for robust feature learning and image transformations. In this paper, we propose an approach that uses recent artistic style transfer algorithms for face sketch synthesis. Additionally, we show that poorly synthesized images can be improved with a denoising autoencoder for better facial feature reconstruction. Further, the approach is extended to perform face verification of heterogeneous image samples to assess the effectiveness of the proposed approach and gives a better view into the potential applications for styling algorithms for face image synthesis and transformation problems alike.