A face recognition algorithm based on multiple convolution kernels and double layer sparse automatic encoder
by Hao Wang; Xiuyou Wang; Huaming Liu; Dongqing Xu; Zhengyan Liu
International Journal of Security and Networks (IJSN), Vol. 15, No. 4, 2020

Abstract: The prevailing face recognition algorithms adopt the manual design feature or the automatic extraction feature of deep learning in the process of characteristic extraction. In order to extract the distinguishing characteristics of the target more accurately, a face recognition algorithm based on the multiple convolution kernels and double-layer sparse automatic encoder was proposed. Initially, the proposed algorithm pre-treated the image with zero-phase component analysis (ZCA) whitening to decrease the correlation of the characteristic and reduced the complexity of the network training. Subsequently, a deep network characteristic extractor was designed, based on the convolution, pooling and multi-layer sparse automatic encoder. The convolution kernel was obtained by an independent unsupervised learning, and an automatic deep characteristic extractor was obtained by pre-training and fine-tuning. Finally, the extracted characteristics were classified using the Softmax regression model. The experiment results manifest that the presented algorithm was superior to the existing algorithms and conventional deep learning algorithms.

Online publication date: Tue, 10-Nov-2020

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Security and Networks (IJSN):
Login with your Inderscience username and password:

    Username:        Password:         

Forgotten your password?


Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com