A Graph-Based Approach for Making Consensus-Based Decisions in Image Search and Person Re-Identification

Area of Science:

• Computer Vision
• Graph Theory
• Artificial Intelligence

Background:

• Image matching and retrieval are crucial in computer vision for applications like image search, biometrics, and person re-identification.
• Existing methods often rely on single algorithms, potentially limiting accuracy and robustness.
• Generating consensus from multiple matching algorithms is a key challenge for improving retrieval performance.

Purpose of the Study:

• To develop a novel algorithm, Shortest Hamiltonian Path Estimation (SHaPE), for generating consensus among multiple image matching and retrieval algorithms.
• To address the problem of ranking candidates based on multiple sets of scores by mapping it to a graph-theoretic problem.
• To demonstrate the efficiency of SHaPE in image search and person re-identification tasks.

Main Methods:

• SHaPE maps the candidate ranking problem to a graph-theoretic problem.
• It incorporates results from multiple sets of scores generated by different matching algorithms.
• A two-step process involving a greedy algorithm and Ant Colony Optimization is used to find a suitable path in the graph.

Main Results:

• Experiments in image search and person re-identification demonstrate the efficiency of SHaPE.
• The algorithm effectively generates consensus among diverse image matching results.
• SHaPE shows promise for improving the accuracy and robustness of image retrieval systems.

Conclusions:

• SHaPE provides an effective method for consensus-based decision-making in image matching and retrieval.
• The algorithm's graph-theoretic approach offers a flexible framework for integrating multiple scoring sources.
• SHaPE's applicability extends beyond image retrieval to any problem requiring ranking based on multiple score sets.