Image steganalysis using active learning and hyperparameter optimization.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Traditional image steganalysis requires extensive labeled datasets, which are costly and time-consuming to create.
• Existing active learning methods for steganalysis often lack flexibility in dynamic environments.

Purpose of the Study:

• To develop an efficient image steganalysis technique that minimizes the need for labeled data.
• To enhance digital security through improved detection of hidden data in images.

Main Methods:

• Combines active learning with off-policy Deep Reinforcement Learning (DRL) for strategic data selection.
• Utilizes the Differential Evolution (DE) algorithm for hyperparameter tuning to ensure model stability.
• Evaluates the approach on the BossBase 1.01 and BOWS-2 datasets.

Main Results:

• Achieves high detection accuracy, with an average F-measure of 93.152% on BossBase 1.01 and 91.834% on BOWS-2.
• Demonstrates robust ability to distinguish between unaltered and steganographic images.
• Confirms improved sample efficiency and learning outcomes due to off-policy DRL.

Conclusions:

• The proposed method effectively enhances image steganalysis accuracy using minimal labeled data.
• The integration of active learning and off-policy DRL offers a flexible and efficient solution for digital security.
• This research contributes a significant advancement in detecting hidden data within digital images.