Age Estimation from Facial Photos: A CNN-Based Approach with Multi-Model Feature Fusion

Zixuan Lyu

doi:10.54097/g24fan25

Authors

Zixuan Lyu

DOI:

https://doi.org/10.54097/g24fan25

Keywords:

age estimation, image recognition, CNN, feature fusion, transformer.

Abstract

This study was designed to improve the accuracy and efficiency of facial age recognition by training and tuning advanced deep learning models. This study evaluated the performance using the UTK Face dataset and multiple deep learning models for the facial age recognition task and found that a three-model fusion (EfficientNetB0, DenseNet121, and Inception V3) had the best accuracy and generalization ability. The study also tried adding a Transformer layer, but the results were not significantly improved. Triple Model Fusion (EfficientNetB0, DenseNet121, and Inception V3) with the addition of the Transformer layer had an overall accuracy of 64%, which was slightly lower than the version without the Transformer (65% accuracy). Triple Model Fusion performed in all tested configurations Best with 65% accuracy.

Downloads

Download data is not yet available.

References

S. Ritz-Timme et al., "Age estimation: the state of the art in relation to the specific demands of forensic practise," Int. J. Legal Med., vol. 113, pp. 129-136, 2000. DOI: 10.1007/s004140050283

J. Devlin, M. W. Chang, K. Lee, and K. Toutanova, "Bert: Pre-training of deep bidirectional transformers for language understanding," arXiv:1810.04805, vol. preprint October 11. Available: https://doi.org/10.48550/arXiv.1810.04805

K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition," in Proc. IEEE Conf. Comput. Vis. Pattern Recogn. (CVPR), Las Vegas, Nevada, 2016, pp. 770-778: IEEE.

A. Vaswani et al., "Attention is all you need," in Proc. 31st Conf. Neural Inform. Process. Syst. (NIPS 2017), Long Beach, CA, USA, 2017, vol. 30: NeurIPS.

C. Szegedy et al., "Going deeper with convolutions," in Proc. IEEE Conf. Comput. Vis. Pattern Recogn. (CVPR), Boston, Massachusetts, 2015, pp. 1-9: IEEE.

K. Simonyan and A. Zisserman, "Very deep convolutional networks for large-scale image recognition," in Proc. 3rd Int. Conf. Learn. Represent. (ICLR 2015), San Diego, 2014: Computational and Biological Learning Society (CBLS).

S. Qiu. (n.d., August 23, 2023). UTKFace large scale face dataset. Available: https://susanqq.github.io/UTKFace/

Skillcate. (n.d., August 24, 2023). Age detection model using CNN: A complete guide. Available: https://medium.com/@skillcate/age-detection-model-using-cnn-a-complete-guide-7b10ad717c60

M. Abadi et al., "TensorFlow: A system for large-scale machine learning," in Proc. 12th USENIX Symp. Operat. Syst. Design Implement. (OSDI ’16), Savannah, GA, USA, 2016, pp. 265-283: USENIX.

X. Zhang, "Attention model ensemble method for cataract diagnosis," master’s thesis, Sichuan University, Chengdu, 2021.

J. Alammar. (2018, August 26, 2023). The illustrated transformer. Available: https://jalammar.github.io/illustrated-transformer/