Spectral Regression based Local Discriminant Embedding Algorithm for Face Recognition

Local discriminant embedding algorithm (LDE) can get better recognition performance than the conventional dimensionality reduction algorithms based on subspaces techniques, but LDE is weak generalization performance for high dimension small sample and has huge workload to decompose dense matrix. In this paper, the SR-LDE algorithm is proposed. The spectral regression method is introduced into LDE to improve its generalization performance and reduce complexity for dense matrix decomposition. The experiments show that SR-LDE algorithm has better performance on recognition rate and computing speed.