Abstract
We propose a criminal suspect face generator based on GANs to replace the problematic process of traditional criminal investigation. The proposed method consists of two crucial components: DCGAN for feature-based face generation, and StyleGAN for feature-guided face manipulation. The proposed framework addresses several practical problems of criminal investigation by first generating a face based on a feature and then transferring the face to different features or expressions for unlimited times in an objective and efficient way. Preliminary experimental results are promising to address the problems of crime investigations in real scenarios.