DarwinGSE: Towards better image retrieval systems for intellectual property datasets

Area of Science:

• Computer Science
• Intellectual Property Law
• Image Processing

Background:

• Graphical trademarks are crucial for brand recognition and require robust protection against infringement.
• Existing Content-based Image Retrieval (CBIR) systems struggle with the complexity and scale of industrial property image datasets.
• Reliable and efficient image retrieval is essential for safeguarding intellectual property.

Purpose of the Study:

• To develop a novel CBIR system that addresses the limitations of current solutions in the industrial property sector.
• To enhance the efficiency and reliability of trademark image searching.
• To provide a scalable solution for managing and retrieving trademark images from large, unlabeled datasets.

Main Methods:

• A modular CBIR system employing multiple, weighted feature descriptions for generalized image representation.
• Utilizing Watershedding K-Means segments for evaluating general features, edge maps, and regions of interest.
• Implementing a new similarity measure for image recovery and incorporating daily updates for current results.

Main Results:

• The proposed system achieves a timely retrieval speed, with 95% of searches completed within 10 seconds.
• Demonstrates a high mean average precision of 93.7%, indicating significant reliability.
• Successfully handles large-scale, unlabeled datasets, overcoming previous limitations.

Conclusions:

• The developed CBIR system offers a practical and effective solution for real-world industrial property protection.
• The system's modular design, advanced feature integration, and efficient similarity measure contribute to its superior performance.
• This advancement supports the ongoing need for robust trademark image searching and copyright infringement defense.