GaussianFace

Implementation of the GaussianFace algorithm for TU Delft IN4393 Computer Vision 2016/2017

The repository does not contain the dataset of images. The data can be places in colorferet/output/ and is expected to be normalized images from the gray FERET dataset of size 150x130pixels.

The code used in the final system can be found in src/gaussianface.py and demos/

To run the system install and create a virtual environment with Python 2.7. Please note that creating the F matrix for a lot of images as required in training and testing takes a long time.

To use precalculated F and W matrices (k=3). Unzip savedMatrix/savedMatrix.zip so that the npy files are in ./savedMatrix/

pip install virtualenv
virtualenv project

Activate and configure virtualenv

source project/bin/activate (Linux)
pip install -r requirements.txt

To run training and testing: (Requires FERET normalized images and takes a long time or the precalculated F matrices) The normalized feret images should be placed in /colorferet/output in pgm format (examplename: 00001fa010_930831.pgm). Train the LDA transformation matrix

python -m demos.trainWLDA 

Test on the testdata

python -m demos.testAlgorithm 

Test the LBP faces

python -m demos.lbpfaces <imgpath>