Forensic face image generation and recognition

Abstract

One of the toughest Heterogeneous face recognition scenarios, involving the comparison of face images residing in different modalities, is face photo-sketch recognition. In forensic science, face sketch will be used for identification of criminals based on the description provided by the eyewitness. With the latest technologies available, the face sketches are obsolete, and they do not accurately support recognition and identification of criminals. In this work, we propose a system that generates the image of the suspect using the novel Transparent Latent Generative adversarial network (TL-GAN) is proposed. The generated image can be compared with images from the police database, using the DeepFace facial recognition system. The performance of the system has been compared with various CNN models and proved to be performing well.

Keywords

• gan
• deepface
• deep learning
• face generation
• face recognition