Scholarlink

<p>Human history, spanning approximately 2 million years, showcases the remarkable evolution of Homosapiens from foragers to modern individuals. Throughout this journey, archeological findings underscore the profound role of artistic expressions like dance and music in shaping human civilizations since ancient times. Today, India boasts a rich tapestry of diverse dance forms, with the <i>Sangeet Natak Akademi</i> recognizing eight traditional styles as classical dances: <i>Bharatanatyam, Kathak, Kathakali, Kuchipudi, Manipuri, Mohiniyattam, Odissi</i>, and <i>Sattriya</i>. Dance, as a performative art, serves as a medium for conveying concepts and emotions through coordinated bodily movements, gestures, and facial expressions. Despite significant advancements in hand gesture recognition, research on facial emotion and body posture recognition within Indian classical dance remains limited. In this domain, the lack of large-scale public datasets has been a major challenge faced by the research community. Hence, this paper proposes a novel dataset focused on analyzing facial expressions <i>(Navarasas)</i> in these dance forms. To enhance the recognition of complex facial expressions represented in Indian classical dance, we developed a customized convolutional neural network (CNN) model. This model achieved reasonably good accuracy in recognizing and classifying the <i>Navarasas</i>. Further analysis includes the use of t-distributed Stochastic Neighbor Embedding (t-SNE) for visualizing both the raw data and the features extracted by the CNN at its final layer, offering clear insights into the model’s ability and the dataset’s resilience to distinguish between different expressions. We conducted a comparative analysis with state-of-the-art models like VGG16, ResNet50, InceptionV3, EfficientNet, Vision Transformer (ViT), and Swin Transformer to evaluate their efficacy in this specialized context. This comparative approach also highlights the <i>Navarasa</i> dataset’s robustness and suitability for capturing the complex expressions of Indian classical dance. Furthermore, this research contributes to the development of tools for automated emotion recognition in performing arts, with implications for both artistic pedagogy and technological innovation. This not only enhances our understanding of the emotional palette of Indian Classical Dance but also paves the way for interdisciplinary research at the intersection of art, culture, and technology.</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

Visual analysis and recognition of facial expressions in Indian classical dance using a novel dataset

  • 期刊论文

  • 发表日期：

    2025年5月7日

查看全文
  • Sajitha V. Raj,
  • T. T. Sunil,
  • R. Sivakumar,
  • R. Pradeep

摘要

Human history, spanning approximately 2 million years, showcases the remarkable evolution of Homosapiens from foragers to modern individuals. Throughout this journey, archeological findings underscore the profound role of artistic expressions like dance and music in shaping human civilizations since ancient times. Today, India boasts a rich tapestry of diverse dance forms, with the Sangeet Natak Akademi recognizing eight traditional styles as classical dances: Bharatanatyam, Kathak, Kathakali, Kuchipudi, Manipuri, Mohiniyattam, Odissi, and Sattriya. Dance, as a performative art, serves as a medium for conveying concepts and emotions through coordinated bodily movements, gestures, and facial expressions. Despite significant advancements in hand gesture recognition, research on facial emotion and body posture recognition within Indian classical dance remains limited. In this domain, the lack of large-scale public datasets has been a major challenge faced by the research community. Hence, this paper proposes a novel dataset focused on analyzing facial expressions (Navarasas) in these dance forms. To enhance the recognition of complex facial expressions represented in Indian classical dance, we developed a customized convolutional neural network (CNN) model. This model achieved reasonably good accuracy in recognizing and classifying the Navarasas. Further analysis includes the use of t-distributed Stochastic Neighbor Embedding (t-SNE) for visualizing both the raw data and the features extracted by the CNN at its final layer, offering clear insights into the model’s ability and the dataset’s resilience to distinguish between different expressions. We conducted a comparative analysis with state-of-the-art models like VGG16, ResNet50, InceptionV3, EfficientNet, Vision Transformer (ViT), and Swin Transformer to evaluate their efficacy in this specialized context. This comparative approach also highlights the Navarasa dataset’s robustness and suitability for capturing the complex expressions of Indian classical dance. Furthermore, this research contributes to the development of tools for automated emotion recognition in performing arts, with implications for both artistic pedagogy and technological innovation. This not only enhances our understanding of the emotional palette of Indian Classical Dance but also paves the way for interdisciplinary research at the intersection of art, culture, and technology.