This source code is the implementation of the paper:

Quoc-Tin Phan, Duc-Tien Dang-Nguyen, Giulia Boato, Francesco G. B. De Natale, "Using LDP-TOP in Video Based Spoofing Detection". (submitted to ICIAP 2017).

Prerequisites

• LibSVM version 3.20
• Matlab R2014a or newer

How to use the code?

Feature extraction (/LDP-TOP)

Define parameters:

• ROI_width: width of region of interest.
• ROI_height: height of region of interest.
• nframe: number of frames used in LDP-TOP calculation.
• filter: type of denoising filter. This variable can take one value of {'median', 'gaussian', 'wiener'}.

Load video and call either:

• LDP_TOP_2nd_hist_ff_noise: using second-order LDP-TOP.

or

• LDP_TOP_3rd_hist_ff_noise: using third-order LDP-TOP.

Examples (take a look at feature_extraction_example.m for more details):

 ROI_width = 256;
 ROI_height = 256;
 nframe = 100;
 Rt = 1;
 from_frame = 1;
 filter = 'gaussian';
 feature_dim = 3*4*256*length(Rt);

 data = read_video('101_0001.MOV')

 hist2 = LDP_TOP_2nd_hist_ff_noise(data, ROI_width, ROI_height, nframe, Rt, from_frame, filter);
 hist3 = LDP_TOP_2nd_hist_ff_noise(data, ROI_width, ROI_height, nframe, Rt, from_frame, filter);

Details of training and testing set on UVAD is located in /UVAD.

Classification (/classification)

After download LibSVM 3.20,

• We implement Histogram Intersection Kernel which is not originally supported by LibSVM. To use the new kernel, replace {svm.c,svm.h} by /libsvm/{svm.c,svm.h} and build LibSVM following the user guide.
• Replace mex files in /libsvm by your mex files built from LibSVM.

Suppose your extracted features are stored in mat files:

• data_train.mat contains:
  • data_train: training features.
  • labels_train: training labels.
• data_test.mat contains:
  • data_test: testing features.
  • labels_test: testing labels.

Before running classification, you need to normalize your features by calling the Matlab script /classification/normalize_data.m. After this step, you should have data_train.scaled and data_test.scaled.

Finally you can run classification by calling /classification/run_classification.m.

You have question?

Please feel free to contact Quoc-Tin Phan quoctin.phan@unitn.it.