Abstract: A feature extraction algorithm based on locality and globality was proposed, which on one hand takes the locality of the data into consideration, on the other hand takes the globality of data into account. Consequently, the data afto dimension reduction not only preserves the locality relationship, but also reconstructs and represents the original dato perfectly. PCA (Principal Component Analysis) can represent the data nicely and LPP (Locality Preserving Projection) can preserve the locality relationship, so the algorithm just hybrid the two of them. But LPP does not utilize the classification information, so first LPP-based algorithm is improved, a supervised version is given, which results in that the algorithm is more suitable for classification task. Experiments in face recognition validate the correctness and effectiveness of the proposed algorithm.

Key words: Feature extraction, Locality, Globality, LPP, PCA, Face recognition