Collaborative projection pursuit for face recognition

This paper introduces a new collaborative feature extraction method based on projection pursuit with application to face recognition. We propose a new projection pursuit index based on the weighted sum of six state of the art indices. Using a genetic search, the hyperparameters of the proposed projection index as well as of the selected classifier were jointly optimized to improve the generalization performance of the model. The characteristics of the proposed projection index confer the extracted features with highly discriminative properties as well as robustness against outliers. Those characteristics were evaluated on the Carnegie Mellon University (CMU) face images dataset and their performance was compared against their corresponding Eigenfaces and Fisherfaces. Our method shows competitive results in terms of generalization performance and dimensionality reduction.