Prototyping and transforming facial textures for perception research

TLDR

Facial image transformation along perceived dimensions such as age, gender, race, or health is used in psychology, medicine, and forensics, yet existing methods struggle especially when altering age. The study reviews current face prototyping and transformation techniques and introduces a novel wavelet‑based method for prototyping and transforming facial textures. The method employs prototype images to define salient features, constructs transformation axes from prototype differences, applies wavelet‑based adjustments to input faces, and mitigates texture loss caused by blending.

Abstract

Transforming facial images along perceived dimensions (such as age, gender, race, or health) has application in areas as diverse as psychology, medicine, and forensics. We can use prototype images to define the salient features of a particular face classification (for example, European female adult or East-Asian male child). We then use the differences between two prototypes to define an axis of transformation, such as younger to older. By applying these changes to a given input face, we can change its apparent age, race, or gender. Psychological investigations reveal a limitation with existing methods that's particularly apparent when changing the age of faces. We relate the problem to the loss of facial textures (such as stubble and wrinkles) in the prototypes due to the blending process. We review the existing face prototyping and transformation methods and present a new, wavelet-based method for prototyping and transforming facial textures.