Beyond view transformation: feature distribution consistent GANs for cross-view gait recognition

Gait recognition systems have shown great potentials in the field of biometric recognition. Unfortunately, the accuracy of gait recognition is easily affected by a large view angle. To address the problem, this study proposes a feature distribution consistent generative adversarial network (FDC-GAN) to transform gait images from arbitrary views to the target view and then perform identity recognition. Besides reconstruction loss, view classification and identity preserving loss are also introduced to guide the generator to produce gait images of the target views and keep identity information simultaneously. To further encourage the network to generate gait images whose feature distribution can well align the true distribution, we also exploit the recently proposed recurrent cycle consistency loss, which can help to remove the unnoticed and useless content preserved in the generated gait images. The experimental results on datasets CASIA-B and OU-MVLP demonstrate the state-of-the-art performance of our model compared to other GAN-based cross-view gait recognition models.