Abstract: Recently, bidirectional two-dimensional Local Preserving Projection（2DLPP） is proposed for face representation and recognition. Compared with two-dimensional Local Preserving Projection（2DLPP）, the main idea behind bidirectional 2DLPP is that bidirectional 2DLPP simultaneously considering the row and column directions of images. Bidirectional 2DLPP needs fewer coefficients for image representation than 2DLPP which essentially works in the row direction of images. Furthermore, to enhance the classification ability of bidirectional 2DLPP, a new supervised algorithm is proposed: bidirectional 2DLPP plus LDA, in which images preprocessed by bidirectional 2DLPP are processed by LDA. Theoretical analyses show that bidirectional 2DLPP plus LDA algorithm has advantages over PCA+LDA, 2DLPP and bidirectional 2DLPP in computation complexity and memory requirements. The results of face recognition experimental on ORL and Yale face databases also show good performance of the methods with less computation.

Key words: discriminant bidirectional two-dimensional local preserving projection, linear discriminant analysis, face recognition, calculation complexity

摘要： 双向二维局部保持映射（双向2DLPP）与二维局部保持映射（2DLPP）比较，双向2DLPP同时对图像的行方向和列方向进行降维处理，可以采用较少的系数有效地表示图像。为了进一步增强双向2DLPP算法的分类能力，将双向2DLPP所提取的特征采用线性判别式分析（LDA）进行分类，从而形成了一种新的监督算法：鉴别双向二维局部保持投影。理论分析表明，无论在计算量还是内存要求方面，所提鉴别双向二维局部保持投影算法比双向2DLPP和主成分分析+线性判别式分析（PCA+LDA）要少，而且在ORL 和Yale数据库上的人脸识别实验表明，新算法的识别性能比双向2DLPP和PCA+LDA算法要好，且具有较少的计算复杂度。

关键词: 双向二维局部保持映射, 线性判别式分析, 人脸识别, 计算复杂度