A Novel Trademark Image Retrieval System Based on Multi-Feature Extraction and Deep Networks

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• Traditional graphical search engines struggle with large, complex image datasets.
• Existing systems face challenges in accurately comparing abstract images and industrial property visuals.

Purpose of the Study:

• To develop a highly accurate image retrieval system for complex and large datasets.
• To enhance image comparison through novel techniques for industrial property applications.

Main Methods:

• A multi-phase approach combining deep learning and image processing techniques.
• Utilizing image signatures with abstraction levels for robust comparison.
• Implementing parallel processing for efficient multi-image searches.
• Employing a new similarity compound formula for comprehensive signature analysis.
• Using deep convolutional networks for semantic feature extraction.

Main Results:

• The developed system achieves high accuracy in image retrieval across diverse image typologies.
• The multi-phase approach effectively handles large datasets and abstract images.
• Combining multiple image assets enhances comparison accuracy.

Conclusions:

• The proposed system offers a significant improvement in image retrieval accuracy and efficiency.
• The novel approach, particularly image signatures and parallel processing, is effective for complex image search tasks.
• Deep learning integration enables semantic understanding for abstract image analysis.