Enhancing Sketch-Based Image Retrieval by Re-Ranking and Relevance Feedback

Area of Science:

• Computer Science
• Information Retrieval
• Artificial Intelligence

Background:

• Sketch-based image retrieval (SBIR) faces challenges balancing efficiency and precision.
• Index structures in large-scale databases can be affected by quantization errors.
• Ambiguity in user sketches degrades traditional SBIR performance.

Purpose of the Study:

• To propose an effective sketch-based image retrieval approach.
• To improve the performance of SBIR systems by addressing efficiency and precision tradeoffs.
• To integrate re-ranking and relevance feedback for enhanced retrieval.

Main Methods:

• Developed an SBIR approach incorporating re-ranking and relevance feedback schemes.
• Utilized semantics from query sketches and initial top-ranked images.
• Applied relevance feedback to refine search results for input query sketches.

Main Results:

• The integrated re-ranking and relevance feedback schemes demonstrated mutual benefits.
• The proposed approach significantly improved the performance of sketch-based image retrieval.
• Addressed limitations of traditional methods and quantization errors in index structures.

Conclusions:

• The proposed SBIR method effectively optimizes the efficiency-precision tradeoff.
• Combining re-ranking and relevance feedback offers a robust solution for SBIR.
• This approach enhances the accuracy and reliability of retrieving images from sketches.