Realistic face manipulation by morphing with average faces

Abstract

Face manipulation has become a standard feature of many social media services. Most of these applications use the feature for entertainment purposes. However, such manipulation techniques could also have potential in a journalistic setting. For instance, one could create realistic, anonymized faces, as an aesthetic alternative to the coarse techniques of blurring or pixelation normally used today. In this paper, we describe how we can use algorithms for face manipulation from computer vision to anonymize faces in journalism. The technique described uses morphing with average faces from a selection of faces that is similar to the original face, and alters the faces in the original pictures into realistic - looking face manipulations. However, it struggles with sufficient anonymization due to identifiable non - facial features of persons in an image.