Scholarlink

<p>Emotion recognition from videos plays a vital role in enhancing interactive experiences across computer graphics and virtual reality (VR) applications. This paper presents EmotionViA, an attention-enhanced CNN-LSTM framework designed for real-time emotion recognition from facial expressions in videos. The framework integrates convolutional neural networks (CNNs) for spatial feature extraction and long short-term memory (LSTM) networks for capturing temporal dynamics, while an attention mechanism selectively emphasizes salient facial regions to improve classification accuracy. To further support research reproducibility, we introduce the EmotionViA dataset, encompassing diverse emotional expressions under varied conditions. Experimental results on EmotionViA, FER-2013, AffectNet, and RAF-DB demonstrate that our method surpasses state-of-the-art approaches in both accuracy and real-time performance. EmotionViA holds potential for immersive applications in education, entertainment, health care, and marketing. Our code is available at <a href="https://github.com/rizwanchouhan/emovid">https://github.com/rizwanchouhan/emovid</a>.</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

Attention-enhanced CNN-LSTM framework for real-time video-based emotion recognition

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月29日

查看全文
  • Anandhavalli Muniasamy,
  • Rizwan Abbas,
  • Galiya Ybytayeva,
  • Ashwag Alasmari,
  • Nouf Aldahwan,
  • Hend Khalid Alkahtani

摘要

Emotion recognition from videos plays a vital role in enhancing interactive experiences across computer graphics and virtual reality (VR) applications. This paper presents EmotionViA, an attention-enhanced CNN-LSTM framework designed for real-time emotion recognition from facial expressions in videos. The framework integrates convolutional neural networks (CNNs) for spatial feature extraction and long short-term memory (LSTM) networks for capturing temporal dynamics, while an attention mechanism selectively emphasizes salient facial regions to improve classification accuracy. To further support research reproducibility, we introduce the EmotionViA dataset, encompassing diverse emotional expressions under varied conditions. Experimental results on EmotionViA, FER-2013, AffectNet, and RAF-DB demonstrate that our method surpasses state-of-the-art approaches in both accuracy and real-time performance. EmotionViA holds potential for immersive applications in education, entertainment, health care, and marketing. Our code is available at https://github.com/rizwanchouhan/emovid.