Skip to Main Content
In this paper, we describe a feature extraction algorithm called discriminant uncorrelated locality aware embedding, DULAM for short, which is based on LPP (locality preserving projection). LPP can preserve the local structure of the data, but does not take the class information into account, besides, the extracted feature might be highly correlated. To overcome these drawbacks, DULAM is proposed, which not only preserves the locality of the data, but also takes the class information into consideration, and an uncorrelated constraint is also imposed to reduce the redundancy, thus it betters the recognition performance. Experiments validate the correctness and effectiveness of the algorithm.