Title: Research on a feature fusion-based image recognition algorithm for facial expression

Authors: Yilihamu Yaermaimaiti; Tusongjiang Kari

Addresses: School of Electrical Engineering, Xinjiang University, Urumqi, Xinjiang 830047, China ' School of Electrical Engineering, Xinjiang University, Urumqi, Xinjiang 830047, China

Abstract: In order to solve the problem that recognition rate of facial expression images is easily affected by non-uniform illumination factors, an improved face recognition algorithm is proposed in this paper. Firstly, a facial expression image with Log-Gabor feature vectors of multiple scales and directions is extracted from a face image, and then all the Log-Gabor feature vectors are blocked in a unified way. Secondly, gist algorithm is applied to extract gist feature blocks from Log-Gabor feature vector image, and then all those blocks are cascaded together as the feature vectors of a face expression sample. The fused feature vectors of the face expression sample are trained as the input feature of the stacked auto-encoder (SAE). Finally, the trained expression features are input into the classifier for recognition to obtain the final recognition result. Whether it is in the facial expression database JAFFE or the Uyghur facial expression database, its facial expression recognition rate is the highest, which verifies the superiority of the algorithm we put out in this paper.

Keywords: feature vectors; feature blocks; stacked auto-encoder; SAE; Uyghur facial expression database; UFED.

DOI: 10.1504/IJICT.2023.128712

International Journal of Information and Communication Technology, 2023 Vol.22 No.2, pp.133 - 146

Received: 03 Dec 2020
Accepted: 02 Feb 2021
Published online: 02 Feb 2023 *

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