Mixtures of Local Linear Subspaces for Face Recognition

Traditional subspace methods for face recognition compute a measure of similarity between images after projecting them onto a xed linear subspace that is spanned by some principal component vectors (a.k.a. \eigenfaces") of a training set of images. By supposing a parametric Gaussian distribution over the subspace and a symmetric Gaussian noise model for the image given a point in the subspace, we can endow this framework with a probabilistic interpretation so that Bayes-optimal decisions can be made. However, we expect that diierent image clusters (corresponding, say, to diierent poses and expressions) will be best represented by diierent subspaces. In this paper, we study the recognition performance of a mixture of local linear subspaces model that can be t to training data using the expectation maximization algorithm. The mixture model outperforms a nearest-neighbor classi-er that operates in a PCA subspace.