Adaptive clustering of image database (ACID) as an efficient tool for improving retrieval in a CBIR system

Area of Science:

• Computer Science
• Information Retrieval
• Machine Learning

Background:

• Content-based image retrieval (CBIR) systems traditionally rely on standard relevance feedback (RF) mechanisms.
• Existing CBIR systems may lack adaptability to individual user preferences and subjective needs.

Purpose of the Study:

• To introduce and evaluate an adaptive clustering of image database (ACID) system for CBIR.
• To enhance the speed and subjectivity of image retrieval by incorporating user feedback adaptively.

Main Methods:

• Implemented an ACID system that clusters relevant images based on user input.
• Representative members of clusters are used for subsequent searches, replacing the entire database.
• Clusters are adaptively updated after each retrieval session to align with evolving user needs.

Main Results:

• The proposed ACID system demonstrates potential for faster and more subjective image retrieval.
• The system effectively embeds retrieval history into the search process.
• Adaptive updating of clusters ensures continued relevance to user's current needs.

Conclusions:

• The ACID system offers an effective alternative to standard relevance feedback in CBIR.
• Adaptive clustering significantly improves the personalization and efficiency of image retrieval.
• The system's performance was validated using standard image datasets (Corel and MIT).