Identification of natural images and computer-generated graphics based on statistical and textural features

Area of Science:

• Digital Image Forensics
• Computer Vision
• Machine Learning

Background:

• Digital image acquisition pipelines can be complex, leading to challenges in distinguishing authentic images from synthetic ones.
• Accurate identification of image origins is crucial for various applications, including digital forensics and content authentication.

Purpose of the Study:

• To propose a novel scheme for identifying natural images and computer-generated graphics.
• To leverage statistical and textural features for robust image classification.

Main Methods:

• Investigated differences between natural and computer-generated images using statistical and textural analysis.
• Extracted 31 dimensions of features for classification.
• Employed LIBSVM (Support Vector Machine) for the classification task.

Main Results:

• Achieved an identification accuracy of 97.89% for computer-generated graphics.
• Achieved an identification accuracy of 97.75% for natural images.
• Demonstrated superior performance compared to existing methods relying solely on statistical or other specific features.

Conclusions:

• The proposed method offers a highly effective approach for discriminating between natural and computer-generated images.
• The technique shows significant potential for practical implementation in identifying image origins.
• The combination of statistical and textural features provides a robust foundation for image authenticity verification.