MythicVision: a deep learning powered mobile application for understanding Indian mythological deities using weight...

Area of Science:

• Computer Science
• Artificial Intelligence
• Cultural Heritage Studies

Background:

• Foreign tourists face challenges understanding Indian deities due to cultural and informational gaps.
• Existing methods for identifying religious iconography lack specialized applications for Indian mythology.

Purpose of the Study:

• To develop a deep learning-driven mobile application, MythicVision, for recognizing and interpreting images of Indian mythological deities.
• To enhance foreign tourists' understanding of India's rich cultural heritage through accessible technology.

Main Methods:

• Training and evaluating four state-of-the-art deep learning models on a custom dataset of 10,970 Indian deity images.
• Implementing a weight-centric decision mechanism using model-wise weights based on test accuracies for improved classification.
• Integrating the framework into an end-to-end web application for user convenience.

Main Results:

• The MythicVision framework achieved a 96% accuracy on the in-house dataset.
• The weight-centric decision mechanism outperformed traditional majority voting in multi-class image classification.
• The application successfully recognizes and classifies real-time Indian deity images.

Conclusions:

• MythicVision effectively bridges the cultural understanding gap for foreign tourists interested in Indian mythology.
• The developed deep learning approach offers a novel and accurate solution for deity image recognition.
• The publicly released application and source code support academic and non-commercial research.