Linear local tangent space alignment with autoencoder

Abstract Linear local tangent space alignment (LLTSA) is a classical dimensionality reduction method based on manifold. However, LLTSA and all its variants only consider the one-way mapping from high-dimensional to low-dimensional space. The projected data may not accurately effectively “represent” original samples. This paper proposes novel linear autoencoder called LLTSA-AE (LLTSA with Autoencoder). proposed divided into two stages. conventional process of viewed as encoding stage, additional important decoding stage used reconstruct data. Thus, makes embedding more effectively. gets recognition rates 85.10, 67.45, 75.40 86.67% handwritten Alphadigits, FERET, Georgia Tech. Yale datasets, which are 9.4, 14.03, 7.35 12.39% higher than that respectively. Compared some improved methods LLTSA, it also obtains better performance. For example, Handwritten Alphadigits dataset, compared ALLTSA, OLLTSA, PLLTSA WLLTSA, by 4.77, 3.96, 7.8 8.6% It shows an effective method.