Abstract:

Aiming at the sensitivity of manifold learning, such as Locality preserving projection (LPP), to noises or outliers, a robust version of LPP, called Robust LPP (RLPP), was proposed. By this algorithm, all the data was firstly evaluated to obtain the possibility of data points as outliers. And then this information was used for neighborhood selection and low-dimensional embedding. The RLPP was used for the representation of facial images. The experiments on JAFFE database showed the effectiveness of this algorithm.

Key words: locality preserving projection, robustness, facial expression recognition