To address the limitations of existing facial expression recognition methods that are restricted to basic expression categories and have difficulty adapting to new expression classes in dynamic scenarios, a composite facial expression recognition method based on incremental learning and multi-teacher knowledge distillation was proposed. An incremental learning framework was introduced to enable adaptive learning of composite expressions. To mitigate the issue of catastrophic forgetting commonly encountered in incremental learning, a selective clustering data replay strategy was designed to identify representative samples from previous classes, thereby maintaining the model’s retention of prior knowledge. Moreover, a confidence-aware multi-teacher knowledge distillation mechanism was developed, in which the predictions of teacher models are adaptively weighted according to their confidence levels with respect to true labels, ensuring efficient knowledge transfer. In addition, hierarchical knowledge distillation in the intermediate feature space was employed to enhance the student model’s fine-grained representation capability. Experimental results on the CFEE and RAF-DB benchmark datasets show that recognition accuracies of 78.52% and 74.23% are achieved, respectively, outperforming existing methods. The findings demonstrate that the proposed approach effectively alleviates catastrophic forgetting and exhibits superior performance in dynamic facial expression recognition tasks.
SongXiaoru, YangJia, GaoSong, et al. Pedestrianre-recognition method based on attention mechanism and multi-scale featurefusion[J]. Science Technology and Engineering, 2022, 22(4) 1526-1533.
[4]
JiangM, YinS. Facial expression recognition based on convolutional block attention module and multi-feature fusion[J]. International Journal of Computational Vision and Robotics, 2023, 13(1): 21-37.
HuMin, HuPengyuan, GePeng, et al. Video expression recognition method based on facial motion unit and temporal attention[J]. Journal of Computer-Aided Design & Computer Graphics, 2023, 35(1): 108-117.
[7]
JiangC S, LiuZ T, WuM, et al. Efficient facial expression recognition with representation reinforcement network and transfer self-training for human-machine interaction[J]. IEEE Transactions on Industrial Informatics, 2023, 19(9): 9943-9952.
DengYaping, WangXin, YiTiantian. Emotion analysis of face image based on multi-layer cross attention fusion network model[J]. Science Technology and Engineering, 2023, 23(3): 1152-1159.
[10]
SunM, CuiW, ZhangY, et al. Attention-rectified and texture-enhanced cross-attention transformer feature fusion network for facial expression recognition[J]. IEEE Transactions on Industrial Informatics, 2023, 19(12): 11823-11832.
KamińskaD, AktasK, RizhinashviliD, et al. Two-stage recognition and beyond for compound facial emotion recognition[J]. Electronics, 2021, 10(22): 1-18.
[13]
NidhiC, VermaB. A lightweight convolutional swin transformer with cutmix augmentation and CBAM attention for compound emotion recognition[J]. Applied Intelligence, 2024, 54(17): 7793-7809.
[14]
UllahS, OuJ, XieY, et al. Compound facial expressions recognition approach using DCGAN and CNN[J]. Multimedia Tools and Applications, 2024, 83(38): 85703-85723.
[15]
WangF Y, ZhouD W, YeH J, et al. Foster: feature boosting and compression for class-incremental learning[C]//European Conference on Computer Vision. Aviv: Springer, 2022: 398-414.
[16]
ZhouD W, WangQ W, YeH J, et al. A model or 603 exemplars: towards memory-efficient class-incremental learning[J]. arXiv Preprint arXiv:2205.13218,2022.
[17]
UllahS, XieY, OuJ, et al. CNN-transformer architecture solution for compound facial expression recognition[C]//9th International Conference on Computer and Communications. Chengdu:IEEE, 2023: 1804-1808.
[18]
RebuffiS A, KolesnikovA, SperlG, et al. Icarl: incremental classifier and representation learning[C]//Proceedings of the lEEE Conference on Computer Vision and Pattern Recognition. Honolulu:IEEE, 2017: 2001-2010.
[19]
LüY, YanY, XueJ H, et al. Relationship-guided knowledge transfer for class-incremental facial expression recognition[J]. IEEE Transactions on Image Processing, 2024, 33: 2293-2304.