DLLE:a Modified Pose Invariant Face Recognition Algorithm
Zhang Pei-ren · Journal of Chinese Computer Systems · 2009
Classical LLE(Locally Linear Embedding) finds manifold with low dimension embedded in high dimensional space.But just like Isomap and Laplacian Eigenmaps,class labels are not used during manifold learning. Using class label information will help find a more meaningful mapping from image space to lower manifold space.And this mapping can handle pose variance during face recognition procedure.Finally,explicit mapping is learned via a linear approximation.High recognition rate is reached on FERET database during pose change from-60 to +60 degree.