This toolchain implements a parts-based face verification using Gaussian Mixture Models (GMM) described in [McCool2009].
The toolchain is detailed as follows:
- The images from the database are aligned according to the specified eye locations.
- In the feature extraction step any local feature algorithm can be used, such as: DCT or LBP.
- The extracted features from the training set are used to train the Universal Background Model (UBM). The algorithm GMM can be used for this purpose.
- A model is enrolled from several features of one identity.
The model depends on those features and in the UBM trained in the last block.
Models are enrolled for both the development and the test set.
- The scoring step is defined as the LLR between the client model and the UBM using the feature vector of a given probe as input.
The algorithm GMM Scoring can be used for this purpose.
- The analysis step integrates scores from the development and the test set.
- McCool, S. Marcel: Parts-based face verification using local frequency bands. Advances in Biometrics. Springer Berlin Heidelberg, 2009. 259-268.