GridFace: Face Rectification via Learning Local Homography Transformations

TL;DR

GridFace is a face recognition method that uses local homography transformations to rectify facial images, reducing geometric variations and enhancing recognition accuracy in unconstrained scenarios.

Contribution

It introduces a novel face rectification approach using local homography transformations learned end-to-end with recognition networks.

Findings

Significant reduction in geometric variations.

Improved face recognition accuracy in unconstrained environments.

Effective regularization based on natural face distribution.

Abstract

In this paper, we propose a method, called GridFace, to reduce facial geometric variations and improve the recognition performance. Our method rectifies the face by local homography transformations, which are estimated by a face rectification network. To encourage the image generation with canonical views, we apply a regularization based on the natural face distribution. We learn the rectification network and recognition network in an end-to-end manner. Extensive experiments show our method greatly reduces geometric variations, and gains significant improvements in unconstrained face recognition scenarios.