Title: Asymmetric multi-period deep residual network face age estimation

Authors: Yilihamu Yaermaimaiti; Yan Tian Xing; Tusongjiang Kari

Addresses: School of Electrical Engineering, Xinjiang University, Urumqi, Xinjiang 830017, China ' School of Electrical Engineering, Xinjiang University, Urumqi, Xinjiang 830017, China ' School of Electrical Engineering, Xinjiang University, Urumqi, Xinjiang 830017, China

Abstract: Aiming at the problem that it is difficult for face age estimation to find features that accurately characterise face age changes, this paper proposes an asymmetric multi-period deep residual network face age estimation model. First, an asymmetric residual block is used to improve the feature extraction ability of the network model. Secondly, a multi-period residual structure is built to solve the optimisation problem caused by network deepening. Then, a shortcut connection that combines pooling and convolution is used to reduce the information loss of the shortcut layer. Finally, the age estimation method combining multi-classification and sequential regression is used to reduce the prediction error. Experimental results on three public datasets showed that the error (MAE/RMSE) of this model is reduced to 2.42/3.44, 3.38/4.69 and 5.17/7.34, respectively, which proved the effectiveness of the proposed algorithm.

Keywords: feature extraction; asymmetric convolution; shortcut connection; information loss; error.

DOI: 10.1504/IJCNDS.2024.137079

International Journal of Communication Networks and Distributed Systems, 2024 Vol.30 No.2, pp.227 - 242

Received: 28 Feb 2023
Accepted: 31 Mar 2023
Published online: 01 Mar 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article