Smart-card-based face verification system: Empirical optimization of system meta-parameters

The optimization of a smart card face verification system (SCFVS) design is a complex task. As the parameter's involved are not independent, the search space is of exponential complexity. In our previous work the optimization task was simplified by setting the parameters of the pre-processing stage of the system to default values. Optimization was performed over those parameters involved in the remaining stages and the filtering stage was fixed. In this paper the study is widened by investigating the effect of image filtering and PCA dimension optimization on the system performance. The study is performed using the XM2VTS database. The results of the experiments show that fixing the number of PCA components by the requirement to retain 95% of the total energy does not result in the optimum system performance. A design strategy where the number of PCA components is optimized, after the careful choice of the right Gaussian filtering parameters, can improve performance. However, such an effort is computationally costly and the results are protocol dependent. An alternative solution that achieves high performance while avoiding such a complexity can be obtained by removing the low-order eigenvectors. This is realized by optimizing the binomial filtering parameters while keeping the rest of the preprocessing stage the same.