Image Processing by Preserving a Manifold Structure

Recently, a number of research efforts have shown that the faceimages possibly reside on a nonlinear sub manifold. Earliersystems failed to discover the underlying structure, if the faceimages lie on a nonlinear sub manifold hidden in the imagespace. This paper proposes a new approach to image analysis(representation and recognition), which explicitly considers themanifold structure. To be specific, the manifold structure ismodeled by a nearest-neighbor graph which preserves the localstructure of the image space. An efficient subspace learningalgorithm for face recognition should be able to discover thenonlinear manifold structure of the face space. A face subspaceis obtained based on Locality Preserving Projections (LPP). It isable to preserve local information giving the best representationand lower error rates. Finally, recognition is successfullyperformed on different images to identify a face.