CONVEX HULL ESCAPE PERTURBATION AT EMBEDDING SPACE AND SPHERICAL BINS COLORING FOR 3D FACE DE-IDENTIFICATION
Abstract
This paper proposes a Convex Hull Escape Perturbation (CHEP) method at Embedding Space to achieve 3D Face De-identification. For better reconstruction of the 3D faces, this paper also proposes the Spherical Bins Coloring (SBC) method to reinstate color lost in the SfS process due to change in vertices number. The top κ-accuracies of faces perturbed by CHEP drop significantly when compared to non-perturbed faces.
Subject
Computer Vision
3D
Facial De-identification
Embedding Space
Deep Learning
Generative Adversarial Network (GAN)
Permanent Link
http://digital.library.wisc.edu/1793/82309Type
Thesis