Genink: A Framework for Tattoo Image Synthesis

Tattoos have long been used as complementary information to support the identification of subjects in forensic investigations. This is mainly due to their discriminative patterns and designs, which are valuable in identifying individuals. However, privacy concerns surrounding the collection of real tattoo databases have hampered the development of robust tattoo recognition systems in the past. With recent advances in generative neural networks, it is possible to generate synthetic tattoos, opening up new research directions. In this paper, we propose GenInk, a framework for the generation of synthetic tattoos based on an existing multimodal generative model. To the best of our knowledge, GenInk enables the generation of the first fully synthetic tattoo database that can be used for training tattoo recognition. Experimental evaluations on real tattoo databases show promising results, i.e., rank-1 identification rates above 95 %, when using the proposed synthetic database for training in an identification task.