Multi-channel face reconstruction system based on sketch features using Conditional Adversarial Networks

Face Reconstruction is an important research area in field of Computer Vision and Artificial Intelligence, and has a wide range of related applications in real life. Previous work primarily focuses on face feature extraction through images of real faces, in contrast, our work reconsiders generating real faces with discernible features based on sketch and outlines, in order to be used for face super-resolution reconstruction and face generation, and ultimately for feature face generation using CAN, such as generating real faces according to the descriptions of suspect's facial features by the police. The research content in this paper includes an algorithm for Asian people's face generation without gender bias, a method for generating feature face data sample through digital image processing, and an innovative CAN architecture. The experiments are based on over 14, 000 face samples, which are manually collected from volunteers. Our results show that our face reconstruction system based on sketch features using CAN has higher levels of authenticity, accuracy, and applicability.