融合3D重建特征的面部表情识别OA

Facial expressions are a fundamental medium for emotional communication and represent a significant research direction in the field of affective computing.To address the challenges of recognition accuracy degradation due to illumi-nation changes,partial occlusions,and pose variations in complex scenarios,this paper proposes a facial expression recog-nition network integrating 3D reconstruction features.The network adopts a multi-branch architecture:a keypoint network and an image network extract facial keypoint information and image information from 2D facial images respectively,while a 3D facial reconstruction network simultaneously obtains expression basis features related to illumination,occlu-sion,and pose during the reconstruction process,effectively compensating for feature loss in 2D images.In the feature fusion stage,a multi-feature fusion strategy is designed:a multi-scale cross-fusion Transformer encoder first fuses the key-point and image information from 2D facial images,then integrates them with the 3D expression basis features to enhance recognition accuracy.Experiments on four public datasets,RAF-DB,AffectNet-7,AffectNet-8,and FERPlus,demonstrate that the proposed method achieves recognition accuracies of 92.01%,67.11%,64.00%,and 91.58%,respectively.Com-pared to existing methods,the proposed approach shows significant advantages in complex scenarios involving occlusions and pose variations.The robustness of the approach is primarily reflected in a strong compensation capability for feature loss caused by partial occlusions,and an excellent invariance to appearance deformations induced by head pose changes.This paper provides an effective solution for facial expression recognition in complex scenarios.