基于L1-范数的二维线性判别分析

摘要

为了避免图像数据向量化后的维数灾难问题,以及增强对野值(outliers)及噪声的鲁棒性,该文提出一种基于L1-范数的2维线性判别分析(L1-norm-based Two-Dimensional Linear Discriminant Analysis, 2DLDA-L1)降维方法.它充分利用L1-范数对野值及噪声的强鲁棒性,并且直接在图像矩阵上进行投影降维.该文还提出一种快速迭代优化算法,并给出了其单调收敛到局部最优的证明.在多个图像数据库上的实验验证了该方法的鲁棒性与高效性.%To overcome the curse of dimensionality caused by vectorization of image matrices, and to increase robustness to outliers, L1-norm based Two-Dimensional Linear Discriminant Analysis (2DLDA-L1) is proposed for dimensionality reduction. It makes full use of strong robustness of L1-norm to outliers and noises. Furthermore, it performs dimensionality reduction directly on image matrices. A rapid iterative optimization algorithm, with its proof of monotonic convergence to local optimum, is given. Experiments on several public image databases verify the robustness and the effectiveness of the proposed method.