Scholarlink

<p>Typically, Human facial emotion and expression play a significant part in day-to-day communication and recognizing them is considered one such formidable tasks in the human-computer interface (HCI) field. Facial expression recognition (FER) plays a pivotal role in human-computer interaction. This study introduces an innovative FER method employing a deep learning model optimized with Frog Leap Optimization (FLO) and an Inference Variational Auto Encoder (IVAE) learning classifier. Input facial images undergo feature extraction via Fuzzy Eigen feature extraction, with the extracted features classified using the IVAE algorithm. The classifier’s performance is enhanced through FLO, which identifies optimal values to adjust the IVAE’s non-linearity. Our method was evaluated on three datasets: FER2013, JAFFE, and CK+, demonstrating superior average accuracy (92.4%, 95.4%, and 97.8%, respectively) compared to existing techniques. This research underscores the potential of optimized deep learning models in advancing FER. Also, the performance of classifier with optimization and without optimization model is compared. From the overall outcome the effectiveness of proposed model over other traditional schemes is proven.</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

Enhanced facial expression recognition via frog leap optimization and inference variational auto encoder

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • C. H. Sumalakshmi,
  • P. Vasuki

摘要

Typically, Human facial emotion and expression play a significant part in day-to-day communication and recognizing them is considered one such formidable tasks in the human-computer interface (HCI) field. Facial expression recognition (FER) plays a pivotal role in human-computer interaction. This study introduces an innovative FER method employing a deep learning model optimized with Frog Leap Optimization (FLO) and an Inference Variational Auto Encoder (IVAE) learning classifier. Input facial images undergo feature extraction via Fuzzy Eigen feature extraction, with the extracted features classified using the IVAE algorithm. The classifier’s performance is enhanced through FLO, which identifies optimal values to adjust the IVAE’s non-linearity. Our method was evaluated on three datasets: FER2013, JAFFE, and CK+, demonstrating superior average accuracy (92.4%, 95.4%, and 97.8%, respectively) compared to existing techniques. This research underscores the potential of optimized deep learning models in advancing FER. Also, the performance of classifier with optimization and without optimization model is compared. From the overall outcome the effectiveness of proposed model over other traditional schemes is proven.