Statistical model of quantized DCT coefficients: application in the steganalysis of Jsteg algorithm

Area of Science:

• Digital image processing
• Statistical modeling
• Information security

Background:

• Digital cameras employ image processing pipelines that quantize Discrete Cosine Transform (DCT) coefficients.
• Existing models often rely on empirical data fitting, lacking a robust mathematical framework.
• Steganalysis, the detection of hidden information, is crucial for information security.

Purpose of the Study:

• To propose a novel statistical model for quantized DCT coefficients based on a mathematical framework of the image processing pipeline.
• To utilize the proposed model for accurate detection of hidden information in JPEG images.
• To evaluate the model's performance in steganalysis, specifically for the Jsteg algorithm.

Main Methods:

• Developed a mathematical framework to model quantized DCT coefficients, avoiding empirical data fitting.
• Formulated hidden data detection as a hypothesis testing problem.
• Derived the most powerful likelihood ratio test for steganalysis and designed a maximum likelihood estimator for embedding rate.

Main Results:

• The proposed statistical model accurately represents quantized DCT coefficients.
• The likelihood ratio test demonstrates high statistical performance for Jsteg steganalysis.
• The maximum likelihood estimator effectively determines the embedding rate.

Conclusions:

• The proposed statistical model provides an accurate representation of DCT coefficients.
• The developed steganalysis methods are effective for detecting hidden data in JPEG images.
• The study highlights the model's utility in assessing the statistical performance of steganalysis techniques.