PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
March 10, 2026Computer Graphics Forum0 citations

Enhanced Visual Exploration and Interactive Retrieval of Pedestrian Images With Incremental User Feedback

View Full Paper
WXWang XiaZFZihang FuZZZihao Zhu

Key Points

  • The aim is to improve pedestrian image retrieval efficiency by integrating user feedback and advanced visualization techniques.
  • Developed a composite visualization for efficient browsing and retrieval.
  • Implemented a cluster-based visualization layout to minimize visual occlusion.
  • Designed a multi-scale, pixel-based view for user guidance during exploration.
  • Introduced a k-fusion re-ranking algorithm to support incremental user feedback.
  • Conducted quantitative experiments and usability studies to assess effectiveness.
  • Improved accuracy in pedestrian recognition due to integrated user feedback.
  • Enhanced visual stability and browsing experience for users.
  • Demonstrated effectiveness of the k-fusion re-ranking algorithm in experiments.

Abstract

Abstract Pedestrian image retrieval is a task that aims to retrieve a pedestrian‐of‐interest from overlapping and non‐overlapping camera views in security scenarios. Although existing person re‐identification (Re‐ID) models have achieved remarkable progress on standard datasets, they still encounter numerous challenges in real‐world deployment. Factors such as viewpoint variations, occlusions and illumination changes often hinder the extraction of stable and robust pedestrian features, resulting in insufficient recognition accuracy and a heavy reliance on manual intervention. To bridge the gap between automated models and practical applications, we develop a visual analysis method for visual exploration and interactive retrieval of pedestrian image collection. Our method integrates a composite visualization that supports efficient visual browsing, retrieval and exploration of candidate targets, and a user‐feedback mechanism that incorporates the automatic model with human insights. In the visualization component, we develop a cluster‐based visualization with an optimized layout to reduce visual occlusion and preserve the user's visual stability. We also design a multi‐scale, pixel‐based view to guide user exploration in the search space. To support incremental user feedback, we propose an extended semi‐supervised learning method by introducing a k‐fusion re‐ranking algorithm. We executed two quantitative experiments to validate the effectiveness of the re‐ranking algorithm and the layout algorithm. Additionally, two case studies by experts and a controlled user study were conducted to demonstrate the usability and effectiveness of our system.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Xia et al. (2026) studied this question.

synapsesocial.com/papers/69af951a70916d39fea4c457https://doi.org/10.1111/cgf.70313
Ask AI
Helpful
Bookmark
Share
View Full Paper