Learning to Trace: Expressive Line Drawing Generation from Photographs

Pacific Graphics

Publication date: October 14, 2019

Naoto Inoue, Daichi Ito, Ning Xu, Jimei Yang, Brian Price, Toshihiko Yamasaki

In this paper, we present a new computational method for automatically tracing high-resolution photographs to create expres-sive line drawings. We define expressive lines as those that convey important edges, shape contours, and large-scale texturelines that are necessary to accurately depict the overall structure of objects (similar to those found in technical drawings) whilestill being sparse and artistically pleasing. Given a photograph, our algorithm extracts expressive edges and creates a cleanline drawing using a convolutional neural network (CNN). We employ an end-to-end trainable fully-convolutional CNN to learnthe model in a data-driven manner. The model consists of two networks to cope with two sub-tasks; extracting coarse lines andrefining them to be more clean and expressive. To build a model that is optimal for each domain, we construct two new datasetsfor face/body and manga background. The experimental results qualitatively and quantitatively demonstrate the effectivenessof our model. We further illustrate two practical applications.