Face recognition with pose variations

摘要

A new efficient two-sage algorithm to compute nonlinear features is described. Its implementation on a neural net with adaptive activation functions that raise the input data to an arbitrary power is described. Its use in face recognition with unknown input poses is presented.