A new approach to image copy detection based on extended feature sets

Area of Science:

• Computer Science
• Digital Forensics
• Image Processing

Background:

• Traditional image copy detection relies on features resistant to attacks, but finding universal features is challenging and often domain-specific.
• Existing methods struggle with the difficulty of identifying globally effective features for image copy detection.

Purpose of the Study:

• To propose a novel framework, the extended feature set, for enhanced image copy detection.
• To address the limitations of conventional methods by generating robust features through simulated attacks.

Main Methods:

• A new framework, the extended feature set, is proposed for image copy detection.
• Virtual prior attacks are applied to copyrighted images to generate novel features for training classifiers.
• The generated features are used to train classifiers to solve the copy-detection problem.

Main Results:

• The proposed approach substantially enhances the accuracy of image copy detection.
• Experiment results demonstrate the effectiveness of the extended feature set in improving detection performance.
• The framework can be integrated with existing copy detection systems to boost their performance.

Conclusions:

• The extended feature set framework offers a more effective approach to image copy detection.
• Simulating attacks to generate features provides a robust solution for identifying image copies.
• This method improves upon conventional techniques by enhancing detection accuracy.