Scholarlink

In the realm of non-verbal communication, facial expressions play a crucial role in decoding human emotions and attitudes. Facial expression recognition and sentiment analysis present challenges when considering cultural variations in emotional expression. Hence, this work utilizes the Gabor Filter Principal Component Analysis MobileNetV2 (GFPCA MobNet) approach to create a robust and reliable model for identifying facial expressions sentiments using the JAFFE dataset. The dataset consists of 213 greyscale photos depicting seven emotions (neutral, happy, sorrow, surprise, anger, disgust, and fear). The dataset is augmented by rotation, scaling, and horizontal flipping to improve diversity and resilience. Feature extraction involves the identification of facial landmarks, and feature selection is performed using the Gabor filter and principal component analysis (PCA), respectively. The MobileNetV2 model is generated to recognize seven different facial expression sentiments by fine-tuning the hyperparameters like learning rate, batch size, and dropout rates using grid search. The model performed with an accuracy of 97.82%. This demonstrates the enhanced proficiency of this approach in detecting facial expressions, making it highly efficient for applications that require high accuracy.

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

GFPCA MobNet: An Approach for Sentiment Analysis on Japanese Facial Expression

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • T. Ramanna,
  • Gayathri Hegde,
  • P. Deepa Shenoy,
  • K. R. Venugopal

摘要

In the realm of non-verbal communication, facial expressions play a crucial role in decoding human emotions and attitudes. Facial expression recognition and sentiment analysis present challenges when considering cultural variations in emotional expression. Hence, this work utilizes the Gabor Filter Principal Component Analysis MobileNetV2 (GFPCA MobNet) approach to create a robust and reliable model for identifying facial expressions sentiments using the JAFFE dataset. The dataset consists of 213 greyscale photos depicting seven emotions (neutral, happy, sorrow, surprise, anger, disgust, and fear). The dataset is augmented by rotation, scaling, and horizontal flipping to improve diversity and resilience. Feature extraction involves the identification of facial landmarks, and feature selection is performed using the Gabor filter and principal component analysis (PCA), respectively. The MobileNetV2 model is generated to recognize seven different facial expression sentiments by fine-tuning the hyperparameters like learning rate, batch size, and dropout rates using grid search. The model performed with an accuracy of 97.82%. This demonstrates the enhanced proficiency of this approach in detecting facial expressions, making it highly efficient for applications that require high accuracy.