Face sketch synthesis: a survey

Abstract

Face sketch synthesis (FSS) has been widely applied to various computer vision tasks, such as criminal detection, information security, digital entertainment, etc. In the past several years, various FSS models with promising performance have been proposed. However, an in-depth understanding of these models in this topic remains lacking. The current survey: i) investigates few models; ii) classifies the models abstractly and monotonously; iii) lacks analysis of existing databases. iv) evaluates models in single evaluation metric. In this paper, we provide a comprehensive survey of the 50 state-of-the-art (SOTA) FSS models. Then we further describe the typical models objectively and analyze the results subjectively. Moreover, we divide these models into two main categories: traditional models and deep learning models. In addition, a novel classification is proposed: coefficient models and regression models. Finally, for the aforementioned problems, we discuss several challenges and highlight some directions of FSS for future research about new database and evaluation strategy.