Scholarlink

This paper presents a hybrid deep learning framework for facial age estimation, integrating Convolutional Neural Networks (CNNs) to extract spatial features and Bidirectional Long Short-Term Memory (BiLSTM) networks to model temporal aging patterns. By leveraging both global facial structures and localized sequential dependencies, the proposed method captures intricate aging trends with greater precision. A learnable weighted fusion mechanism dynamically optimizes the contributions of CNN and BiLSTM predictions, enhancing robustness to occlusions and variations in image quality. Unlike fixed fusion strategies, this adaptive approach ensures dataset-specific optimization, improving age estimation accuracy. These findings highlight the effectiveness of multi-scale spatial analysis and bidirectional temporal modeling, making the approach well-suited for applications in biometrics, healthcare, and human-computer interaction.

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

Adaptive Spatial-Temporal Fusion for Robust Facial Age Estimation Using CNN-BiLSTM Architecture

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • P. Elamukilan,
  • Sadhvik Bathini,
  • Santhoshkumar Peddi,
  • Debasis Samanta

摘要

This paper presents a hybrid deep learning framework for facial age estimation, integrating Convolutional Neural Networks (CNNs) to extract spatial features and Bidirectional Long Short-Term Memory (BiLSTM) networks to model temporal aging patterns. By leveraging both global facial structures and localized sequential dependencies, the proposed method captures intricate aging trends with greater precision. A learnable weighted fusion mechanism dynamically optimizes the contributions of CNN and BiLSTM predictions, enhancing robustness to occlusions and variations in image quality. Unlike fixed fusion strategies, this adaptive approach ensures dataset-specific optimization, improving age estimation accuracy. These findings highlight the effectiveness of multi-scale spatial analysis and bidirectional temporal modeling, making the approach well-suited for applications in biometrics, healthcare, and human-computer interaction.