Shuffle dense networks for biometric authentication

Abstract

In recent years, with the continuous development of artificial intelligence technology, biometric authentication based on computer vision technology have also been developed rapidly. In this paper, we propose a novel Shuffle dense networks (SDN) with combining ShuffleNet and DenseNet for biometric authentication. ShuffleNet is an extremely computation-efficient structure that can obtain more channels information. DenseNet makes full use of all hierarchical features, which can facilitate the flow of information. Specifically, dense skip connections is adopted for combining the low-level features and the high-level features to enhance the performance of the reconstruction and residual learning is applied for easing the difficulty of training the deep neural network. In addition, the grouped convolutions are introduced for reducing computational complexity and the number of parameters. What’s more, the shuffle dense connections are proposed for mitigating the grouped convolutions problem of lacking information exchange between the groups. The proposed method is evaluated quantitatively and qualitatively on four benchmark datasets, and the experimental results of super-resolution illustrated that our SDN achieves great performance over the state-of-the-art frameworks.