Evaluation of Eigenfaces and Fisherfaces using Bray Curtis dissimilarity metric

The researchers of biométrie community have built a variety of benchmarks to evaluate face recognition methods. It is vital for researchers to leverage these methods and conduct sound experimental validation along to the comparison with state-of-the-art methods. In past few decades, the Eigenfaces and Fisherfaces face recognition methods are evaluated using the Euclidean distance metric that have shown better recognition accuracy in constrained environments, but insufficient to handle the unconstrained environments such as variations in pose, facial expression, and illumination. This paper presents an evaluation of the traditional face recognition methods such as Eigenfaces and Fisherfaces using Bray Curtis dissimilarity metric in unconstrained environments. The Bray Curtis is a statistical measure of dissimilarity between feature vectors that has the property to retain the values between 0 and 1 for measuring the two extremities of match and mismatch. The normalization is done using absolute difference divided by the summation. It views the space as grid similar to city-box distance. The classification performance of these methods is critically evaluated using Euclidean distance and Bray Curtis dissimilarity metrics in the face recognition problem under different conditions on publicly available face databases such as AT & T-ORL, Indian face database, and extended Yale B. The experimental results show significant improvement of recognition accuracies of Eigenfaces and Fisherfaces methods in the case of extreme variations of illumination when computed using Bray Curtis dissimilarity metric.