A flexible and effective linearization method for subspace learning

In the past decades, a large number of subspace learning or dimension reduction methods [2,16,20,32,34,37,44] have been proposed. Principal component analysis (PCA) [32] pursues the directions of maximum variance for optimal reconstruction. Linear discriminant analysis (LDA) [2], as a supervised algorithm, aims to maximize the inter-class scatter and at the same timeminimize the intra-class scatter. Due to utilization of label information, LDA is experimentally reported to outperform PCA for face recognition, when sufficient labeled face images are provided [2].