Protecting the Privacy of Face by De-Identification Pipeline Based on Deep Learning

In this research, we offer a face de-identification process that is reversible and alters both the face’s geometry and its texture. There are fourteen parameters that are employed for the change of the geometry. The usage of a fixed face texture template is required for texture change. We have explored how the capacity of people and robots to identify faces is affected when different geometrical and surface adjustments are made to components of the face such as the eyes, eyebrows, nose, and lips. The studies using crowdsourcing and machine face recognition were carried out on photographs of well-known persons that were obtained from the internet. The findings collected from both kinds of research indicated that alterations to a person’s facial geometry (form) are less influential on the amount of privacy protection than changes to that person’s face texture.