Image steganalysis feature selection based on the improved Fisher criterion.

Area of Science:

• Digital Image Forensics
• Information Security
• Computer Science

Background:

• Steganalysis relies on feature selection for accurate hidden message detection.
• Traditional Fisher criterion can discard useful feature components, reducing detection accuracy.

Purpose of the Study:

• To propose an improved Fisher criterion (I-Fisher) for enhanced steganalysis feature selection.
• To develop a feature selection method utilizing the I-Fisher criterion and rough set reduction.
• To optimize high-dimensional steganalysis features for improved detection accuracy and reduced dimensionality.

Main Methods:

• Introduced a sigmoid function into Fisher's criterion to create the I-Fisher criterion.
• Applied the I-Fisher criterion as a heuristic function for decision rough set α-positive region reduction.
• Implemented a feature selection method based on the I-Fisher criterion for steganalysis.

Main Results:

• The I-Fisher criterion improves the measurement of steganalysis feature component separability.
• The proposed method effectively reduces the dimension and memory requirements of GFR and CC-PEV features.
• Detection accuracy is maintained or improved using the proposed feature selection method.

Conclusions:

• The I-Fisher criterion-based feature selection method enhances steganalysis performance.
• This approach offers a viable solution for optimizing steganalysis features in digital image forensics.
• The method effectively balances feature dimensionality reduction with detection accuracy.