基于Fisher准则的二次线性特征提取

Secondary Linear Feature Extraction Based on Fisher Criterion

将基于Fisher准则的鉴别分析推广到矩阵的情形,得到先寻找最佳鉴别向量,将矩阵投影为一个向量,再次寻找最佳鉴别向量,将投影向量映射为一个点的二次线性特征提取方法,从而为解决高维、奇异情况下最优鉴别向量集的求解问题提供了新的途径.在ORL标准人脸库上的试验结果表明,该方法识别结果可靠,准确率高.

The method of secondary linear feature extraction is derived by generalizing the linear discrimination analysis to matrix based on Fisher criterion in this present paper. The derivation is to find optimal discrimination vectors for projecting an image matrix to a vector, and then to a point. A new approach for the solution of optimal discrimination vectors under high- dimensional and singular conditions is presented. The classification results are dependable with high accuracy in test of ORL face image database.