Distance Metrics for Face Recognition by 2d Pca

Abstract

Two-dimensional Principal Component Analysis (2D PCA) is a global feature extraction method for Face recognition that works upon 2D matrices rather than 1D vectors. In every Face recognition system, different distance functions used in the classification stage can yield diverse recognition rates and one of the quests for the developers is to figure out which is the most preferable function. In this paper, we concentrate on the insights of distance metrics applied for 2D PCA. A new distance metric so called weighted p, in which an exponent p and eigenvalues are used, is also proposed. To evaluate the recognition performance of those functions, comparative experiments on the face database ORL are performed. The results show that the proposed function provides 2D PCA with higher recognition rates than existing rivals.

Parallel Architecture for Face Recognition using MPI

“…Training and recognition require finiding a relation between these points that enables classification or clustering based on a distance metric like euclidean distance, cosine angle distance, mean square error (MSE) distance, Manhattan distance or correlation distance [15]. Finding such a relation is not easily done and may be erroneous.…”