ABSTRACT
A major kind of face recognition methods is appearance based; taking the whole image as a highdimensional random vector, and applying linear subspace transform method or manifold learning method for feature extraction. There are many linear subspace methods applied in dimension reduction of face image such as principle component analysis (PCA), linear discriminate analysis (LDA), independent component analysis (ICA), and much more improved nonlinear algorithms are developed in recent years. The linear structure of data can be found by the linear subspace method, but the problem is that there is more nonlinear information contained in the data. Studies have shown that the high dimensional face images exist substantially in a lower dimensional manifold [1-2]. Facial images under expression, illumination and pose variation can be reduced to corresponding low-dimensional features with concrete physical meaning, which show good result, so we consider using manifold learning algorithms for face recognition.
