Scholarlink

<p>The traditional CBIR systems that are heavily dependent on low, level features of the image for information retrieval have been criticized for low accuracy and low transparency. As per the latest requirements, such systems need to be equipped with models that generate explanations for the images, which makes the models more interpretable and thus, users can be more assured of the information they receive. The enhanced CBIR system serves as the foundation for the improved Secure Clothing Recommendation Scheme, an advanced image recommendation technique that uses a new hybrid approach that combines RAG descriptors and deep learning-based image feature extraction to provide a more comprehensive retrieval of image content through combined global and region-based feature extraction. Additionally, by including an Explainable AI (XAI) framework into the recommendation process, we provide text and graphics that clarify how and why recommendations were made in order to improve comprehension of the underlying content. In order to prevent unauthorized parties from viewing or using the users’ data, our model uses an encrypted mechanism for exchanging picture features, protecting both the images and the users’ associated data. The results of testing our new Hybrid CBIR scheme on a standard fashion model image dataset demonstrate that our proposed Enhanced version of CBIR is statistically superior to the conventional CBIR systems based on precision, recall, and retrieval efficiency. The new CBIR scheme showed a statistically measurable increase of approximately 12% in accuracy and approximately 15% increase in retrieval time over the existing CBIR systems. The suggested system is a strong and innovative way to overcome the drawbacks of traditional Content, Based Image Retrieval (CBIR) methods. By effectively combining hybrid algorithms with RAG descriptors and Explainable AI, the system manages to enhance accuracy, security, and interpretability all at the same time. This work sets up the base for creating more dependable and user, friendly recommendation systems that are typical of the modern e, commerce sector.</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

A secure encrypted CBIR and explainable hybrid deep learning framework for clothing recommendation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月6日

查看全文
  • Akanksha Mrinali,
  • Pankaj Gupta,
  • Yeswanth Dintakurthy,
  • Arun Kumar Soma,
  • K. R. K. V. Prasad,
  • Jawad Khan,
  • Mohammed Wasim Bhatt

摘要

The traditional CBIR systems that are heavily dependent on low, level features of the image for information retrieval have been criticized for low accuracy and low transparency. As per the latest requirements, such systems need to be equipped with models that generate explanations for the images, which makes the models more interpretable and thus, users can be more assured of the information they receive. The enhanced CBIR system serves as the foundation for the improved Secure Clothing Recommendation Scheme, an advanced image recommendation technique that uses a new hybrid approach that combines RAG descriptors and deep learning-based image feature extraction to provide a more comprehensive retrieval of image content through combined global and region-based feature extraction. Additionally, by including an Explainable AI (XAI) framework into the recommendation process, we provide text and graphics that clarify how and why recommendations were made in order to improve comprehension of the underlying content. In order to prevent unauthorized parties from viewing or using the users’ data, our model uses an encrypted mechanism for exchanging picture features, protecting both the images and the users’ associated data. The results of testing our new Hybrid CBIR scheme on a standard fashion model image dataset demonstrate that our proposed Enhanced version of CBIR is statistically superior to the conventional CBIR systems based on precision, recall, and retrieval efficiency. The new CBIR scheme showed a statistically measurable increase of approximately 12% in accuracy and approximately 15% increase in retrieval time over the existing CBIR systems. The suggested system is a strong and innovative way to overcome the drawbacks of traditional Content, Based Image Retrieval (CBIR) methods. By effectively combining hybrid algorithms with RAG descriptors and Explainable AI, the system manages to enhance accuracy, security, and interpretability all at the same time. This work sets up the base for creating more dependable and user, friendly recommendation systems that are typical of the modern e, commerce sector.