Geometric Coding for Billion-Scale Partial-Duplicate Image Search

Description:

The invention leverages geometric coding to encode spatial relationships among local features of images, enabling the retrieval of partial duplicates through global verification. By utilizing geometric square coding and geometric fan coding to create three geo-maps, this approach significantly enhances the detection of duplicate images altered by rotation, scale changes, occlusion, and background clutter.

Key Advantages:

• High computational efficiency enabling real-time response
• Effective detection of partial duplicates with varied alterations
• Scalable to billion-scale image databases
• Reduces feature quantization loss and embeds geometric consistency
• Improves precision of image retrieval with geometric verification

Problems Solved:

• Difficulty in detecting partial-duplicate images among billions of web images
• Limited efficiency and discriminative power of existing local feature quantization methods
• Expensive computational cost of full geometric verification on large candidate sets
• Trade-offs between feature quantization and geometric verification accuracy in large-scale databases

Market Applications:

• Copyright violation detection for digital images
• Image-based information retrieval systems
• Online content monitoring and management
• Enhanced search engines for partial-duplicate image identification
• Assistance in digital forensics and image-based investigations