Skip to content

liuyang12/privacy-dual-imaging

Repository files navigation

Privacy Dual Imaging (PDI)

This repository contains the MATLAB code for the paper Imaging Privacy Threats from an Ambient Light Sensor (preprint [submitted] 2023) by Yang Liu, Gregory W. Wornell, William T. Freeman, Fredo Durand. [project] [github]

Usage

Prerequisites

  • MatConvNet for FFDNet image denoiser. We highly recommend compiling with GPU support. Note that you might find this issue useful if you compile it on MATLAB 2018a and above.

Note: If you do not have MatConvNet setup already, set USE_MATCONVNET=false in test_privacy_dual_imaging.m and it will use TV denoiser instead.

Run PnP-QCS on experimental data

  1. Test proposed inversion algorithm PnP-QCS on measured pointing hand data (Figure 3) in test_privacy_dual_imaging.m.

  2. All touch gesture figures in the paper can be reproduced by running corresponding scripts in figures/ directory. All data are measured from the Samsung View2 tablet (17.3-inch) using .

Acquire your own data

  • [Android Devices] We use the AndroSensor app made by Fiv Asim to acquire the raw data from the ambient light sensor. We set the "Update Interval" to "Very Fast" to maximize the sensor sampling rate. The raw data is stored in .csv files in the rawdata/ directory.
  • [Mac Devices] We write a new app AmbientLight to acquire the raw data from the ambient light sensor on MacOS (Macbooks and iMacs). Private API of accessing ambient light sensors on macOS hiden by Apple T1/T2 security chips is borrowed and appreciated from DarkModeBuddy by Guilherme Rambo. More details about the app can be found in the AmbientLight repo.

Structure of directories

directory description
AmbientLight Light sensor acquisition app on MacOS
rawdata Raw .csv files acquired from ambient light sensors on Android tablet or Mac laptops / desktops.
dataset Processed data from raw datapoints for each measurement
figures MATLAB scripts to reproduce the results and figures in the paper
packages Proposed PnP-QCS algorithm and other adapted state-of-art algorithms
results results of reconstruction (after reconstruction)
utils utility functions

Citation

@article{Liu23PDI,
   author    = {Liu, Yang and Wornell, Gregory W. and Freeman, William T. and Durand, Fr{\'e}do},
   title     = {Imaging Privacy Threats from an Ambient Light Sensor},
   journal   = {preprint [submitted]},
   year      = {2023},
   type      = {Journal Article}
}

Contact

Yang Liu, MIT CSAIL