Scholarlink

<p>Endoscopy is the most reliable method for detecting and treating gastrointestinal (GI) disorders, although the accuracy of this procedure relies on the ability of the endoscopist. Research on precise automated image retrieval has the potential to reduce cancer incidence, mortality, false positives, and diagnosis time. This study pre-classifies wavelet transforms and guided filter (WTGF)-based enhanced images using ResNet50 and combined color and texture features with ResNet50 features, utilizing correlation-based techniques such as Discriminant Correlation Analysis (DCA) and Canonical Correlation Analysis (CCA). The user query was addressed by retrieving the top K relevant images from the significant class of the pre-classified dataset using a matching algorithm based on fused feature vectors. The query was reformulated using pseudo-relevance feedback, based on the relevancy of the first top K relevant images, and a pertinent set of top K images was acquired. The third set of top K images was formed by combining two previous sets of appropriate photos, and the optimized top K images with the highest precision were evaluated using an ensemble approach. The suggested system outperforms comparable systems, achieving 96.36% and 95.37% mAP in the CCA concatenation mode for the top 10 and top 20 scopes, respectively.</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

GITIRNet: Ensemble-Based Gastrointestinal Tract Image Retrieval Network Using Deep Classification and Query Expansion with Fusion of Traditional and Deep Features

  • 期刊论文

  • 发表日期：

    2026年7月18日

查看全文
  • Mathana Gopal Arulsamy,
  • Pallikonda Rajasekaran Murugan,
  • Arunprasath Thiyagarajan,
  • Kottaimalai Ramaraj

摘要

Endoscopy is the most reliable method for detecting and treating gastrointestinal (GI) disorders, although the accuracy of this procedure relies on the ability of the endoscopist. Research on precise automated image retrieval has the potential to reduce cancer incidence, mortality, false positives, and diagnosis time. This study pre-classifies wavelet transforms and guided filter (WTGF)-based enhanced images using ResNet50 and combined color and texture features with ResNet50 features, utilizing correlation-based techniques such as Discriminant Correlation Analysis (DCA) and Canonical Correlation Analysis (CCA). The user query was addressed by retrieving the top K relevant images from the significant class of the pre-classified dataset using a matching algorithm based on fused feature vectors. The query was reformulated using pseudo-relevance feedback, based on the relevancy of the first top K relevant images, and a pertinent set of top K images was acquired. The third set of top K images was formed by combining two previous sets of appropriate photos, and the optimized top K images with the highest precision were evaluated using an ensemble approach. The suggested system outperforms comparable systems, achieving 96.36% and 95.37% mAP in the CCA concatenation mode for the top 10 and top 20 scopes, respectively.