Scholarlink

<p>In digital archives and cultural preservation, effective image retrieval from extensive datasets is critical. This study develops an efficient Content-Based Image Retrieval (CBIR) system for retrieving culturally significant Indian paintings, enhancing digital accessibility and preservation. It focuses on diverse styles like Gond, Pichwai, Madhubani, Mandla, and Warli. Traditional CBIR relies on handcrafted features, which are time-consuming and less effective. This study proposes a fusion approach using EfficientNetB0 and ResNet50, to enhance retrieval accuracy. Further, Recursive Feature Elimination (RFE) optimizes dimensionality reduction. Various distance metrics, including Hamming, Minkowski, cosine similarity, Jaccard, Euclidean, and Manhattan, compute similarity scores. Performance is assessed using precision, recall, and F1-score. The mean average values of precision, recall, and F1-score of the proposed CBIR system on the TIAPD dataset come out to be 91.64%, 18.17%, and 29.54% respectively. This research not only enhances retrieval accuracy but also contributes to the digital preservation of India’s artistic heritage.</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 digital preservation of Indian art heritage through fusion based deep learning CBIR system

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月24日

查看全文
  • Meenu Garg,
  • Sonam Aggarwal,
  • Ashok Kumar,
  • Rajat Kapila

摘要

In digital archives and cultural preservation, effective image retrieval from extensive datasets is critical. This study develops an efficient Content-Based Image Retrieval (CBIR) system for retrieving culturally significant Indian paintings, enhancing digital accessibility and preservation. It focuses on diverse styles like Gond, Pichwai, Madhubani, Mandla, and Warli. Traditional CBIR relies on handcrafted features, which are time-consuming and less effective. This study proposes a fusion approach using EfficientNetB0 and ResNet50, to enhance retrieval accuracy. Further, Recursive Feature Elimination (RFE) optimizes dimensionality reduction. Various distance metrics, including Hamming, Minkowski, cosine similarity, Jaccard, Euclidean, and Manhattan, compute similarity scores. Performance is assessed using precision, recall, and F1-score. The mean average values of precision, recall, and F1-score of the proposed CBIR system on the TIAPD dataset come out to be 91.64%, 18.17%, and 29.54% respectively. This research not only enhances retrieval accuracy but also contributes to the digital preservation of India’s artistic heritage.