Scholarlink

<p>This study investigates facial emotion recognition in individuals with Down syndrome (DS) using Action Units (AUs) extracted with OpenFace 2.0 and evaluated with ML (Decision Tree, KNN, SVM) and DL (FCNN, 1D-CNN) models. We introduce an intensity-aware AU selection method that identifies a compact and informative subset, and show that preserving moderate-intensity cues is critical for DS. A DS-trained 1D-CNN achieves 94.98% overall accuracy, outperforming an FCNN (90.59%) and ML baselines. Cross-domain tests reveal asymmetry: the DS-trained model generalizes well to the CK+ (typically developing) cohort (96.28%), whereas a CK+-trained model performs poorly on DS (61.48%). The resulting DS-tailored pipeline provides a compact and robust baseline for inclusive emotion recognition in DS.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Emotion Recognition in Individuals with down Syndrome Based on Microexpression Analysis Using Machine Learning and Deep Learning Methods

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Gonzalo Olmedo,
  • Nancy Paredes,
  • Gustavo Simbaña

摘要

This study investigates facial emotion recognition in individuals with Down syndrome (DS) using Action Units (AUs) extracted with OpenFace 2.0 and evaluated with ML (Decision Tree, KNN, SVM) and DL (FCNN, 1D-CNN) models. We introduce an intensity-aware AU selection method that identifies a compact and informative subset, and show that preserving moderate-intensity cues is critical for DS. A DS-trained 1D-CNN achieves 94.98% overall accuracy, outperforming an FCNN (90.59%) and ML baselines. Cross-domain tests reveal asymmetry: the DS-trained model generalizes well to the CK+ (typically developing) cohort (96.28%), whereas a CK+-trained model performs poorly on DS (61.48%). The resulting DS-tailored pipeline provides a compact and robust baseline for inclusive emotion recognition in DS.