Abstract: A novel linear discriminant analysis implementation method is presented to overcome the shortcomings of the traditional LDA algorithm after researching the existing theoretical results. This method amends the Fisher criterion at first, then finds the best discriminant vectors by iteration, analyzes and evaluates them at last. The experimental results on the facial expression recognition using the JAFFE expressions library and the comprehensive evaluation of regional consumption levels show that the PSO-LDA algorithm not only has well recognition effect, but also can break through the restriction of the sample dimension. Compared with other improved LDA algorithm, it’s more flexible, and easier to implement.

Key words: Linear Discriminant Analysis(LDA), projection vector, matrix of the discrete degree, Particle Swarm Optimization(PSO), PSO-LDA , algorithm

摘要： 针对传统线性判别分析方法存在的问题，在研究现有理论成果的基础上，提出一种新的LDA实现方法。该方法首先对原有的Fisher准则进行修正，然后通过迭代搜寻最佳鉴别矢量，最后对获取的鉴别矢量进行比较分析。在标准的JAFFE人脸库上的表情识别和地区综合消费水平的评价中的实验结果表明，此算法不仅具有良好的识别效果而且还可以突破样本维数的限制；与其他LDA算法相比，该算法更具灵活性且更易于实现。

关键词: 线性判别式分析, 投影矢量, 离散度矩阵, 粒子群算法, PSO-LDA算法