Progressive Pose Normalization Generative Adversarial Network for Frontal Face Synthesis and Face Recognition Under Large Pose

This paper proposes a Progressive Pose-Normalization Generative Adversarial Network (PPN-GAN) for frontal face synthesis and face recognition. The key idea is to normalize a profile face progressively: starting from inferring an intermediate face that has a small view difference to the profile face, and then increasing the view difference step by step, until the frontal view of the profile face is recovered. In addition to the progressive strategy, an additional identity discriminator and identity-aware losses in both the image and feature spaces are also incorporated into the GAN for identity preserving. Experimental results show that our method not only produces compelling perceptual results but also outperforms the state-of-the-art methods on face recognition under large-pose.

1. INTRODUCTION

Face recognition is an enduring hot topic in the computer vision community. Although great progress has been made by recent deep neural networks [1], [2], [3], pose variations still challenge face recognition in many practical applications. To address this challenge, two main categories of methods, i.e., pose-specific methods [4], [5] and pose-normalization methods [6], [7], have been proposed. Pose-specific methods integrate multiple or tree-structured models that are specific for different face poses, while pose-normalization methods try to synthesize a frontal face from a profile and then use the synthesized image for identification. Compared to pose-specific methods, pose-normalization methods not only have achieved better performance on many benchmark datasets [6], [8] but also have additional advantages such as intuition, interpretability, and convenience.