This is an application of our methodology to 3D facial imaging data. Due to the sensitive nature of the data, we work with 5000 simulated faces. Our goal is to highlight the difficulty in sanitizing an object as complicated as a face and
to further highlight the role smoothing plays in producing sanitized estimates. Each face is viewed as a 2D manifold sitting in R3 and is sampled densely at 7150 points. Technically, we estimate the RKHS smoothing mean and its sanitized version and see how the different parameters can affect the privacy and utility of the sanitized version.