Applying GANs for Image Synthesis and Recognition in Forensic Contexts

Forensic investigations usually depend on manually drawn facial sketches that are based on witness descriptions, which can result into inaccuracies and time delays. This paper proposes a method of automating face sketch generation using Generative Adversarial Networks (GANs), specifically StackGAN, through deep learning. The system uses Natural Language Processing (NLP) techniques to extract relevant facial features from witness descriptions before creating high- resolution facial sketches using a two-phase StackGAN model. To enhance precision, sketch generation is matched with a criminal database using Convolutional Neural Networks (CNNs) for suspect identification. The proposed method was tested on wellknown public datasets like FFHQ and CelebA, showing noticeable improvements in how realistic the sketches looked and how accurately they resembled real faces. This approach can make forensic investigations more efficient by cutting down on human error, speeding up the process of identifying suspects, and offering a scalable, AI-powered tool for law enforcement.