Local Linear Discriminant Analysis Using \( \ell_{2} \)-Graph

A recently proposed method, called Local Fisher Linear Discriminant Analysis (LLDA), the experiment showed that compared with the traditional Fisher Linear Discriminant Analysis, it has a better result. However, it uses Euclidean distance selecting nearest neighbor samples which has some flaws in the way, such as the robustness is not good and not sparse, and so on. The paper presents an improved approach, called Local Linear Discriminant Analysis Using \( \ell_{2} \)-Graph (L2G_LLDA). It remains reconstructed coefficient of samples to select the nearest samples, which enhances the robustness of the algorithms and makes it sparse. The extensive experimental results over several standard face databases have demonstrated the effectiveness of the proposed algorithm