基于WT-2DICA与改进Fisher算法的人脸识别

Face Recognition Based on WT-2DICA and Improved Fisher Algorithm

小波变换与二维独立元分析(WT-2DICA)能有效提取人脸图像的高阶统计信息,但不能很好地识别受污损的人脸图像。改进Fisher算法充分考虑了类别信息,避免了传统Fisher算法造成的小样本问题。该文结合2种算法的优点,融合改进Fisher算法的最佳投影方向与WT-2DICA算法的独立基子空间,获得了融合投影方向。实验结果表明,该融合算法具有较好的分类性能。

High-order statistical information can be extracted effectively with Two-dimensional Independent Component Analysis based on Wavelet-transform（WT-2DICA）, but the method is not valid in the recognition of the damaged images. Improved Fisher method avoids small samples problem in traditional Fisher method by considering category information. Combined with the advantages of two algorithms, fusion projection direction is obtained, which integrates best projection direction from improved Fisher and independent basis subspace from WT-2DICA. Experimental results show that the fusion method is valid in face recognition.