Year: 2024
Type: conference-paper
Abstract: LBP is an excellent method that effectively describes the local texture of an image. It is not only computationally simple, but also has a good effect, and thus is widely applied in the field of facial expression recognition. However, the original LBP method suffers from weaknesses of poor feature description and generalization ability. To overcome the above mentioned shortcomings, a novel facial expression recognition algorithm that combines improved ResNet model and LBP feature extraction method (LBP-ResNet) is proposed in this paper. First, the gradient boosting decision tree algorithm is adopted to locate the key facial points, cropping the facial region images to reduce the interference from non-expression areas. Secondly, the facial image size is normalized to extract the LBP feature maps, and data augmentation is performed through affine transformation. These feature maps are then input into an improved ResNet network, which introduces squeeze and excitation attention modules and replaces the original ReLu static function with DyReLu dynamic function to enhance the features of key facial points. Finally, the LBP-ResNet algorithm is compared with several representative algorithms on CK+ and JAFFE datasets. Experimental results demonstrate that the proposed algorithm shows a better performance in facial expression recognition accuracy and generalization ability.
Authors: Fang Dong, Jianhui Zhong, Wenfeng Wang, Jia Han, Tao Chen
Language: English
Topic: Emotion and Mood Recognition
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: bronze
OpenAlex ID: W4412100074