Multiple Query Content-Based Image Retrieval Using Relevance Feature Weight Learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Information Retrieval

Background:

• Content-based image retrieval (CBIR) systems often struggle with the semantic gap between low-level visual features and high-level user semantics.
• Existing retrieval methods may not effectively adapt feature relevance to specific user queries.

Purpose of the Study:

• To propose a novel multiple query retrieval approach, named weight-learner.
• To reduce the semantic gap in image retrieval by learning optimal relevance weights for visual features.

Main Methods:

• The weight-learner approach uses visual feature discrimination to estimate image distances.
• It employs unsupervised learning to determine optimal relevance weights for each visual descriptor per query image.
• The solution is mathematically formulated through the minimization of objective functions to optimize feature weights.

Main Results:

• The proposed method successfully learns feature relevance weights tailored to user queries.
• Weight-learner demonstrated effectiveness in reducing the semantic gap, leading to improved image retrieval accuracy.
• The approach was validated using an image collection from the Corel database.

Conclusions:

• The weight-learner approach offers a promising solution for enhancing multiple query image retrieval.
• Learning optimal feature relevance weights is crucial for bridging the semantic gap and improving CBIR performance.