Scholarlink

<p>Affective Computing is a concept born to endow the computational machine with the understanding of human emotions, and this study introduces the Adaptive Vision Enhancement Suite (AVES), a comprehensive pipeline for facial emotion recognition, combining image preprocessing techniques with the DeepFace library. The framework enhances the image quality through adaptive methods such as Contrast Limited Adaptive Histogram Equalization (CLAHE), Gaussian Filtering and Gamma Correction, depending on the quality attributes of the image, asserted via metrics such as energy, sharpness, and contrast. Emotion classification and confidence scores were derived from both raw and enhanced images, enabling a comparative analysis of preprocessing effects. In the AVES Suite, Statistical tests, including paired t-tests and Wilcoxon signed-rank tests, demonstrated significant improvements in the classification confidence post-enhancement. Dimensionality reduction (t-SNE, PCA) and K-means clustering facilitate emotion pattern visualization and grouping. An Emotion Quality Metric was introduced, which integrates confidence scores with quality measures, providing an end-to-end measure of emotion detection performance. Evaluated on multiple public datasets including FER-2013 and CK + to demonstrate the generalizability of the proposed method, AVES improved classification in 39–56% of cases, with up to an 11.4% accuracy gain on 1000 samples. The method is especially effective for noisy, low-resolution inputs commonly found in real-world applications such as driver monitoring, remote learning, customer sentiment analysis, and telemedicine. By enhancing image quality prior to classification, AVES improves the robustness and reliability of emotion recognition systems.</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

Enhancing DeepFace algorithm performance for emotion detection: an adaptive vision preprocessing approach using FER-2013 dataset

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Navyamol K. T.,
  • Rubin Thottupurathu Jose

摘要

Affective Computing is a concept born to endow the computational machine with the understanding of human emotions, and this study introduces the Adaptive Vision Enhancement Suite (AVES), a comprehensive pipeline for facial emotion recognition, combining image preprocessing techniques with the DeepFace library. The framework enhances the image quality through adaptive methods such as Contrast Limited Adaptive Histogram Equalization (CLAHE), Gaussian Filtering and Gamma Correction, depending on the quality attributes of the image, asserted via metrics such as energy, sharpness, and contrast. Emotion classification and confidence scores were derived from both raw and enhanced images, enabling a comparative analysis of preprocessing effects. In the AVES Suite, Statistical tests, including paired t-tests and Wilcoxon signed-rank tests, demonstrated significant improvements in the classification confidence post-enhancement. Dimensionality reduction (t-SNE, PCA) and K-means clustering facilitate emotion pattern visualization and grouping. An Emotion Quality Metric was introduced, which integrates confidence scores with quality measures, providing an end-to-end measure of emotion detection performance. Evaluated on multiple public datasets including FER-2013 and CK + to demonstrate the generalizability of the proposed method, AVES improved classification in 39–56% of cases, with up to an 11.4% accuracy gain on 1000 samples. The method is especially effective for noisy, low-resolution inputs commonly found in real-world applications such as driver monitoring, remote learning, customer sentiment analysis, and telemedicine. By enhancing image quality prior to classification, AVES improves the robustness and reliability of emotion recognition systems.