An Efficient Method for Facial Sketches Synthesization Using Generative Adversarial Networks

Abstract

The synthesis of facial sketches is an important technique in digital entertainment and law enforcement agencies. Recent advancements in deep learning have shown its possibility in generating images/sketches using attribute guided features. Facial features are important attributes because they determine human faces' detailed description and appearance during sketch generation. Traditionally, the forensic or composite artist has to sketch by interviewing witnesses manually. To automate this process of face sketch generation, a deep learning-based generative adversarial network incorporated with multiple activation functions is proposed for its efficiency improvement. The proposed model is extensively tested using different evaluation metrics such as RMSE, PSNR, SSIM, SRE, SAM, UIQ & BRISQUE.