Abstract:
"
The “forensic face generation” is one of the major fields in forensic science that helps
criminal investigations to carry out their investigation process. According to a survey
conducted by United States Law Enforcement Agencies confirms that 80% of agencies
use computer-automated systems while Sri Lanka is still far behind in the process of
face generation with a lot of inefficiencies in the current manual process. Hence this
research introduces a novel approach for the manual face generation process, while
eliminating the inefficiencies of the manual procedure of Sri Lanka. In order to
overcome this situation, this study introduces an automated deep learning-based
software solution targeting the Sri Lankan population. In this research, an image
synthesis has been done according to a given feature by using StyleGAN. The
controlled image was retrieved by using a correlation between the features of generated
image and the noise in latent space. By this approach, when a one facial attribute is
given as input, the generated image gives an accuracy of 79.28%. The ultimate goal of
this research study is to provide a system for law enforcement agencies to carry out an
efficient and effective face generation process that can lead to an increase the success
rate of suspect identification.
"
