Introducing set of internal parameters for Laplacian faces to enhance performance under varying conditions

Laplacianfaces is a recent addition to appearance based face recognition algorithms with promising future potential. Unlike Eigenfaces algorithm, Laplacianfaces algorithm finds an embedding that preserves the locality information of the subjects in feature space. In this study we have comprehensively evaluated the performance of Laplacianfaces against PCA on FERET face-image database using csuFaceIdEval as the testing platform. The effect of internal parameters, including size of locality to be preserved, the choice of distance measure to determine locality and the number of leading eigenvalues to be used for matching has been thoroughly studied for the first time. The impact of illumination, face expression and age variations on the relative performance of Laplacianfaces and Eigenfaces has been shown to be very significant and best parameter settings for enhanced performance have been proposed.