Robust Discriminant Projection Via Joint Margin and Locality Structure Preservation

摘要

It is very challenging to obtain sufficiently discriminative features from the original data in real-world applications. Despite the multiplicity of researches on the linear discriminative analysis, most of them are sensitive to noise, outliers and the distribution of data, especially in the low sample size context. In this paper, we propose a novel image classification method, namely Margin and Locality Discriminant Projection, which simultaneously considers the margin and locality structure information based on low-rank and sparse representation. Specifically, the proposed method integrates marginal fisher analysis and neighborhood preserving embedding so as to preserve the intrinsic structure as well as enhance the discriminative ability, on account of which a more robust and comprehensive graph can be constructed to obtain sufficiently discriminative features. Meanwhile, the low-rank and sparsity constraints are introduced to compensate the noise. The proposed model can be solved efficiently using the linear alternative direction method with adaptive penalty and eigen-decomposition. Extensive experiments are conducted on four databases and the results demonstrate that the proposed method can achieve superior performance than other state-of-the-art algorithms.