Learning local convolutional features for face recognition with 2D-warping

Abstract

The field of face recognition has seen a large boost in performance by applying Convolutional Neural Networks (CNN) in various ways. In this paper we want to leverage these advancements for face recognition with 2D-Warping. The latter has been shown to be effective especially with respect to pose-invariant face recognition, but usually relies on hand-crafted dense local feature descriptors. In this work the hand-crafted descriptors are replaced by descriptors learned with a CNN. An evaluation on the CMU-MultiPIE database shows that in this way the classification performance can be increased by a large margin.