Kenny V. dos Santos, Luiz Eduardo S. e Silva and Waldir S. S. Junior, Federal University of Amazonas, Brazil

Discriminative filtering is a pattern recognition technique which aim maximize the energy of output signal when a pattern is found. Looking improve the performance of filter response, was incorporated the principal component analysis in discriminative filters design. In this work, we investigate the influence of the quantity of principal components in the performance of discriminative filtering applied to a facial fiducial point detection system. We show that quantity of principal components directly affects the performance of the system, both in relation of true and false positives rate.

Pattern Recognition, Discriminative Filtering, Principal Component Analysis & Fiducial Points Detection.