Sketch to image synthesis: harnessing deep learning techniques for realistic visual transformation

Crime poses a significant threat to society, with rising rates in developing countries often attributed to unidentified criminals and various social factors. Traditional black-and-white sketches used by law enforcement agencies pose challenges as witnesses struggle to definitively identify suspects. In order to enhance the accuracy of identification of criminals from face sketches, we propose a deep learning based generative model that can convert face sketches into realistic colour images. The model we have proposed consists of Component Embedding (CE), Feature Mapping (FM) and Image Synthesis (IS) modules which work together to produce realistic images through the use of sophisticated feature learning techniques. The CE module captures key facial components, while FM and IS modules work in tandem to synthesise high-quality colour images from sketches. The proposed approach is evaluated on the CelebA-HQ dataset and shows better performance in both qualitative and quantitative comparison with previous methods. The contribution of this paper can help to advance the field of sketch-to-image synthesis and provide a strong solution for the problem of criminal identification challenges.