Privacy-Preserving Annotation of Face Images through Attribute-Preserving Face Synthesis

Shirai, Sola

Student Work

Privacy-Preserving Annotation of Face Images through Attribute-Preserving Face Synthesis

Public

In this project we investigate the viability of collecting annotations for face images while preserving privacy by using synthesized images as surrogates. We compare two approaches: a deep learning model to render a detailed 3D reconstruction of the face from an input image; and a novel generative adversarial network architecture that extends BEGAN-CS to generated images conditioned on desired facial features. Using these two models, we conduct an experiment with crowdsourced workers to compare annotation quality of original face images and synthesized versions. Across 60 workers annotating a total of 180 images (60 of each version), we find that while original versions have the best accuracy (84.5%), the 3D (75.9%) and GAN (75.6%) versions show promising results.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator