Indexed by:
Abstract:
Facial expressions carry the emotional state of human beings and are one of the most powerful signals for humans to express themselves. Therefore, the research of facial expression recognition has very important significance and broad application prospects. However, in complex scenes, the blurring of facial images, the interference of redundant information, and the similarity of expressions will cause unsatisfactory expression recognition. In response to these shortcomings, this article adds a convolutional attention module to the basic Resnet network structure, infers the attention map in turn along two independent dimensions (channel and space), performs adaptive feature optimization, captures expression feature information, and reduces human faces. The interference of redundant information; and the introduction of island loss ISlandloss classification to optimize the distance between classes, so that the network can learn more discriminative features, and improve the discrimination of expressions. Compared with the existing mainstream facial expression classification algorithms, this method has certain advantages in objective evaluation indicators. The experimental results on FER2013 show that the accuracy of this algorithm is 73.85%, which is higher than 71.23% of the Resnet network. © Published under licence by IOP Publishing Ltd.
Keyword:
Reprint Author's Address:
Email:
Source :
ISSN: 1742-6588
Year: 2021
Issue: 4
Volume: 1952
Language: English
Cited Count:
SCOPUS Cited Count:
ESI Highly Cited Papers on the List: 0 Unfold All
WanFang Cited Count:
Chinese Cited Count:
30 Days PV: 7
Affiliated Colleges: