Exploring machine learning approaches for efficient image forgery detection

Area of Science:

• Computer Science
• Digital Forensics
• Image Processing

Background:

• Accessible image manipulation tools raise concerns about visual content authenticity.
• Forgery techniques threaten personal, journalistic, and security contexts, necessitating robust detection methods.
• Maintaining trust in digital media requires reliable systems to identify altered images.

Purpose of the Study:

• To develop a robust system for detecting various image forgeries (copy-move, splicing, object removal).
• To minimize false positives and negatives in image forgery detection.
• To enhance machine learning models' ability to recognize and identify forged images.

Main Methods:

• Utilized the Two-dimensional maximum Shannon Entropy Median Filter (TSETMF) for noise reduction and detail enhancement.
• Employed Multidimensional Spectral Hashing (MSH) for efficient, compact feature extraction and improved pattern recognition.
• Implemented Faster Region-Based Convolutional Neural Networks (FR-CNN) for rapid localization and feature extraction of manipulated areas.

Main Results:

• The proposed machine learning approach achieved high accuracy (98.5%), precision (97.0%), recall (98.2%), and F1 score (98.1%).
• The integrated system demonstrated robustness against diverse forgery methods.
• Real-time analysis capabilities were enhanced through improved processing speed and accuracy.

Conclusions:

• Machine learning techniques, including CNNs and MSH, significantly improve image forgery detection accuracy, speed, and robustness.
• The developed system effectively combats evolving image forgery methods.
• Future research should focus on more robust models, unsupervised learning, and cross-domain applications.