antispoofing.fusion 1.0.0

Using an automatic installer

Using pip is the easiest (shell commands are marked with a $ signal):

$ pip install antispoofing.fusion

You can also do the same with easy_install:

$ easy_install antispoofing.fusion

This will download and install this package plus any other required dependencies. It will also verify if the version of Bob you have installed is compatible.

This scheme works well with virtual environments by virtualenv or if you have root access to your machine. Otherwise, we recommend you use the next option.

Using zc.buildout

Download the latest version of this package from PyPI and unpack it in your working area. The installation of the toolkit itself uses buildout. You don’t need to understand its inner workings to use this package. Here is a recipe to get you started:

$ python bootstrap.py
$ ./bin/buildout

These 2 commands should download and install all non-installed dependencies and get you a fully operational test and development environment.

Finding Valid Output Scores

The previously generated outputs do not contain a valid score for each video frame. The motion based countermeasure needs 20 video frames for analyzing the motion correlation between the face and the background, i.e. the method cannot produce scores for the first 19 frames. On the other hand, the LBP based countermeasure is able to produce an valid output score when a face is successfully detected and the face size is above 50x50 pixels.

Therefore, the frames, in which the both countermeasures have valid score (i.e. not NaN value), must be found before performing the fusion at score level. This process is performed using the script ./bin/find_valid_frames.py and by giving the locations of all used output scores, e.g.:

$ ./bin/find_valid_frames.py -s scores/motion_lda scores/lbp_lda -e replay

Combining the Valid Output Scores

The script fuse_scores.py performs the fusion of the any countermeasures at score level using some of the two different methods: sum of scores or logistic linear regression (LLR) with the selected score normalization scheme: minmax, zscore or without any normalization, e.g.:

$ ./bin/fuse_scores.py -s scores/motion_lda scores/lbp_lda -f SUM -n ZNorm -o scores/lda_sum_z
$ ./bin/fuse_scores.py -s scores/motion_lda scores/lbp_lda -f LLR -n None -o scores/lda_llr_raw

Analyzing the Results of Fusion at Frame-level

The performance of the individual countermeasures and their fusion can be dumped in to a file ./results/frame_based/results.txt using the script frame_by_frame_analysis.py:

$ ./bin/frame_by_frame_analysis.py -s scores/motion_lda scores/lbp_lda -f scores/lda_sum_z scores/lda_llr_raw -e replay

The results.txt shows the performance of each method at frame-level.

Running the Time Analysis

The time analysis is the end of the processing chain, it fuses the scores of instantaneous scores for each method to give out a better estimation of attacks and real-accesses. To use it:

$ ./bin/time_analysis.py -s scores/motion_lda scores/lbp_lda -f scores/lda_sum_z scores/lda_llr_raw -e replay

The time evolution for each method can be found in directory ./results/evolution/. The folder also contains a PDF file in which you can find all methods in same figure.

Mutual Error Analysis

The script venn.py performs mutual error analysis on the given countermeasures and outputs the results into a file ./results/Venn&scatter/Venn.txt:

$ ./bin/venn.py -s scores/motion_lda scores/lbp_lda -e replay