Facial Recognition of Sketch Images in Forensic Laboratories Employing Diverse Techniques

The use of Differential Facial Recognition (DFR) over the past few years emerged as a challenging endeavor within the realms of biometrics and computer vision, grappling persistently with the complexities of illumination and pose variations. This scholarly investigation aims to propose innovative deep-learning architectures tailored to juxtapose non-visible facial depictions against an array of visible facial galleries. The taxonomy of thermal-to-visible recognition delineates into two distinct categories: feature-based methodologies and image synthesis paradigms. Notably, the latter enhances compatibility with existing recognition frameworks in both commercial and governmental sectors, bolstering efficacy in forensic examination. Additionally, the incorporation of soft biometrics, encompassing diverse traits such as age and gender, provides supplementary layers of information, thereby reinforcing the foundation of recognition algorithms. Novel strategies are introduced to navigate the intricate landscape of auxiliary training data in the LUPI scenario, pushing the boundaries of recognition performance. Additionally, a pioneering aggregation framework is conceived to enhance the robustness of landmark detection, while adversarial techniques amplify the efficacy of landmark detection mechanisms. Finally, this study scrutinises the opaque veil enveloping Generative Adversarial Networks (GANs), aiming to address concerns regarding mode collapse and diversity within the GAN framework.