基于扰动方法和广义K-L变换的人脸特征抽取

Feature Extraction Based on Perturbation Method and Generalized K-L Transformation

对广义最佳鉴别矢量的求解方法进行研究,根据矩阵的扰动理论和广义K-L变换,提出了一种改进的求解广义最佳鉴别矢量集的解析算法。由于该算法一次性地求出了所有广义最佳鉴别矢量,而无需迭代,因而节约了计算时间,且识别率高。在ORL人脸数据库的数值实验,验证了上述论断的正确性。

A study was made on the solving method of the generalized optimal set of discriminant vectors. An improved analytical algorithm of the generalized optimal set of discriminant vectors was proposed based on the combination of perturbation theory and the K-L transformation. The computational time was saved because all the discriminant vectors were obtained simultaneously with the proposed algorithm, while it did not need iteration. Furthermore, the new algorithm yielded high recognition rate. These statements are supported by the numerical simulation experiments conducted on ORL face database.