WNet: Multi-Task Artistic Portrait Generation Method

Artistic portrait usually uses a set of sparse continuous lines to capture a person's unique appearance characteristics based on real people or photos. This paper addresses the issues of strong background noise and the loss of key details of human faces in artistic portraiture, leading to the inability to produce realistic and high quality portraits. A multi-task portrait painting model with hierarchical loss called WNet is proposed based on U<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>Net used in salient object detection. This model can generate highly abstract portrait painting that do not depend on the background. Compared to existing popular portrait painting generation methods, the images generated by our method have a better visual effect and retain the key details of the facial features better. The experimental results indicate that the proposed WNet is optimal in terms of PSNR, SSIM and MAE performance. Finally, The software communicating robotic arm can quickly spot-paint the fused art portraits.