LBP特征和改进Fisher准则的人脸识别

为了进一步提高人脸识别系统的性能，在LDRC算法的基础上进行改进，并将改进LDRC算法的准则函数应用到Fisher分类器中，提出了一种新的基于LBP特征和改进Fisher准则的人脸识别算法。该算法提取每幅人脸图像的标准LBP直方图特征：把提取到的LBP特征输入到改进后的Fisher分类器中，得到最佳投影矩阵和投票结果矩阵；求解出投票结果矩阵的最大值所对应的类别号，将其作为最终的识别结果；分别在FERET和AR人脸库中进行实验检测，结果表明与传统的特征提取方法相比，给出的方案可以使人脸识别率得到显著提高。

Face recognition based on LBP feature and improved Fisher criterion

In order to improve the performance of face recognition system, the criterion function of LDRC algorithm is improved and applied to the Fisher classifier, a new face recognition algorithm based on LBP feature and improved Fisher criterion is proposed. Firstly, the standard LBP histogram feature of each face image is extracted. Secondly, the extracted LBP features are input into the improved Fisher classifier to obtain the optimal projection matrix and the voting result matrix. Thirdly, the class number corresponding to the maximum value of the voting result matrix is solved, which is regarded as the final recognition result. Finally, those experiments are done in FERET and AR face database respectively. The final results show that the face recognition rate has been significantly improved compared with the traditional methods.