PCA based face recognition and testing criteria

In this work, we use the PCA based method to build a face recognition system with a recognition rate more than 97% for the ORL and 100% for the CMU databases. However, the main goal of this research is to identify the characteristics of face recognition rates while, i) the number of training and test data is varied; ii) the amount of noise in the training and test data is varied; iii) the level of blurriness in the training and test data is varied; iv) the image size in the training and test data is varied; and v) different databases are used with aligned images. We have observed that, i) in general the increase of the number of signature on images increases the recognition rate, however, the recognition rate saturates after a certain amount of increase; ii) the increase in the number of samples used in the calculation of covariance matrix increases the recognition accuracy for a given number of individuals to identify; iii) the increase in noise and blurriness affects the recognition accuracy; iv) the reduction in image-size has very minimal effect on the recognition accuracy; v) if less number of individuals are supposed to be recognized then the recognition accuracy increases; and vi) aligned images used increases the recognition accuracy.