Anti-Spoofing Descriptor based on Visual Rhythm

Brief Description

This program compute the noise information contained in a input video and their respective spectrum fourier, generating a fourier spectrum video. After, the program compute the visual rhythm from fourier spectrum video, summarizing the most important information for face spoofing detection in a single image, named as visual rhythm.

Requirements

The OpenCV Package 2.4.10 or later is required for use this software. This software runs on Linux Operating systems, and we tested it by using Ubuntu 14.04 LTS and Debian 8.

Compiling and Running this Software

To compile this software the use needs set the variables defined in opencv.inc:

OPENCVFLAGS = -I/path/to/include
OPENCVLIBS = -L/path/to/lib

For example, suppose you have installed Opencv Package in the /home/user/local/opencv-2.4.10, then the OPENCVFLAGS and OPENCVLIBS variables should be set as:

OPENCVFLAGS = -I/home/user/local/opencv-2.4.10/include
OPENCVLIBS = -L/home/user/local/opencv-2.4.10/lib

After set the variables into ./Release/opencv.inc, the user must be run the follows commands to compile the source code:

make -C Release clean
make -C Release

First command line remove old binaries, and the second command builds a new binary named as ./Release/VisualRhythmAntiSpoofing.

How to Use this Software?

This software run only by command line interfaces (CLIs) such as the shell program (e.g., sh, bash, ksh). We provide the following parameters to the users that can be setted by the command line:

• color_space: Integer between 0 and 1 that indicates the color space used to load the video frames (default=0). Use:

  • 0: To load the frames in grayscale;
  • 1: To load the frames in the Lab color space;

• filter: Integer between 0 and 1 that indicates the type of filter used to compute the residual noise video (default=0). Use:

  • 0: To use a median filter;
  • 1: To use a gaussian filter.

• frame_number: Positive integer that indicates the number of consecutive frames used during computation of the visual rhythm (default=50).

• input_video: Filename of the input video to be computed the visual rhythm <required>.

• kernel_size: Positive odd integer that indicates the size of the kernel used during filtering of the input video (default=7).

• output_image: Filename of the computed visual rhythm. Visual rhythm is saved as PNG image file <required>.

• roi_width: Positive integer that indicates the width of the region of interesting extracted of each frames (default=30).

• variance: Float that indicates the variance of the Gaussian filter (default=2).

• visual_rhythm_type: Integer between 0 and 2 that indicates the type of visual rhythm to be computed from input video <required>. Use:

  • 0: To compute a vertical visual rhythm;
  • 1: To compute a horizontal visual rhythm;
  • 2: To compute a zig-zag visual rhythm.

P.S.: The parameters must be setted using a hyphen (-) before the name of the parameter followed by blanck space and their value (e.g., -visual_rhythm_type 0, -frame_number 50).

Examples

For run the sofware the user must be the follow parameters:

1. Compute the vertical visual rhythm of an input video (EXAMPLE/data/testcase1.avi) using default value parameters:

./Release/VisualRhythmAntiSpoofing -visual_rhythm_type 0 -input_video EXAMPLE/data/testcase1.avi -output_image EXAMPLE/output/visualrhythm/vertical/testcase1.png

2. Compute the vertical visual rhythm (-visual_rhythm_type 0) from first 2 seconds (-frame_number 50) of an input video (EXAMPLE/data/testcase2.avi):

./Release/VisualRhythmAntiSpoofing -visual_rhythm_type 0 -frame_number 50 -color_space 0 -roi_width 30 -filter 0 -kernel_size 7 -variance 2 -input_video EXAMPLE/data/testcase2.avi -output_image EXAMPLE/output/visualrhythm/vertical/testcase2.png

3. Compute the horizontal visual rhythm (-visual_rhythm_type 1) from first 2 seconds (-frame_number 50) of an input video (EXAMPLE/data/testcase3.avi):

./Release/VisualRhythmAntiSpoofing -visual_rhythm_type 1 -frame_number 50 -color_space 0 -roi_width 30 -filter 0 -kernel_size 7 -variance 2 -input_video EXAMPLE/data/testcase3.avi -output_image EXAMPLE/output/visualrhythm/vertical/testcase3.png

4. Compute the zig-zag visual rhythm (-visual_rhythm_type 2) from first 2 seconds (-frame_number 50) of an input video (EXAMPLE/data/testcase1.avi):

./Release/VisualRhythmAntiSpoofing -visual_rhythm_type 2 -frame_number 50 -color_space 0 -roi_width 30 -filter 0 -kernel_size 7 -variance 2 -input_video EXAMPLE/data/testcase1.avi -output_image EXAMPLE/output/visualrhythm/vertical/testcase1.png

Please, Cite our Work!

If you use this software, please cite our paper published in IEEE Transactions on Information Forensics and Security:

Reference

Pinto, A.; Robson Schwartz, W.; Pedrini, H.; De Rezende Rocha, A., "Using Visual Rhythms for Detecting Video-Based Facial Spoof Attacks," Information Forensics and Security, IEEE Transactions on , vol.10, no.5, pp.1025,1038, May 2015
doi: 10.1109/TIFS.2015.2395139
keywords: {Authentication;Biometrics (access control);Databases;Face;Feature extraction;Histograms;Noise;Unicamp Video-Attack Database;Video-based Face Spoofing;Video-based face spoofing;Visual Rhythm, Video-based Attacks;impersonation detection in facial biometric systems;unicamp video-attack database;video-based attacks;visual rhythm},
URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7017526&isnumber=7073680

Bibtex

@ARTICLE{7017526,
author={Pinto, A. and Robson Schwartz, W. and Pedrini, H. and De Rezende Rocha, A.},
journal={Information Forensics and Security, IEEE Transactions on},
title={Using Visual Rhythms for Detecting Video-Based Facial Spoof Attacks},
year={2015},
month={May},
volume={10},
number={5},
pages={1025-1038},
keywords={Authentication;Biometrics (access control);Databases;Face;Feature extraction;Histograms;Noise;Unicamp Video-Attack Database;Video-based Face Spoofing;Video-based face spoofing;Visual Rhythm, Video-based Attacks;impersonation detection in facial biometric systems;unicamp video-attack database;video-based attacks;visual rhythm},
doi={10.1109/TIFS.2015.2395139},
ISSN={1556-6013},}

License

This software is available under condition of the AGPL-3.0 Licence.

Copyright (c) 2015, Allan Pinto, William Robson Schwartz, Helio Pedrini, and Anderson Rocha