Revision history [back]

Since facerecognizer is implemented as a demo face recognition (FR) system to show that a real FR application is viable, it does not have good performance (speed and accuracy) with large training sets or real life applications. You can see that the author uses PCA for eigenvalues, eigenvectors computations, and a lot of matrix multiplication, which are not advantages of OpenCV. You can get faster training stage if using other libraries which are dedicated (or just better than OpenCV) for matrix multiplication and eigen computation. Armadillo and Eigen are some preferable choices.

Since facerecognizer is implemented as a demo face recognition (FR) system to show that a real FR application is viable, it does not have good performance (speed and accuracy) with large training sets or real life applications. You can see that the author uses PCA for eigenvalues, eigenvectors computations, and a lot of matrix multiplication, which are not advantages of OpenCV. You can get faster training stage if using other libraries which are dedicated (or just better than OpenCV) for matrix multiplication and eigen computation. Armadillo Armadillo and Eigen Eigen are some preferable choices.