Toward agent-based LSB image steganography system

Abstract

In a digital communication environment, information security is mandatory. Three essential parameters used in the design process of a steganography algorithm are Payload, security, and fidelity. However, several methods are implemented in information hiding, such as Least Significant Bit (LBS), Discrete Wavelet Transform, Masking, and Discrete Cosine Transform. The paper aims to investigate novel steganography techniques based on agent technology. It proposes a Framework of Steganography based on agent for secret communication using LSB. The most common image steganography databases are explored for training and testing. The methodology in this work is based on the statistical properties of the developed agent software using Matlab. The experiment design is based on six statistical feature measures, including Histogram, Mean, Standard deviation, Entropy, Variance and Energy. For steganography, an Ensemble classifier is used to test two scenarios: embedding a single language message and inserting bilingual messages. ROC Curve represents the evaluation metrics. The result shows that the designed agent-based system with 50% training/testing sample set and 0.2 Payload can pick out the best cover image for the provided hidden message size to avoid visual artifact.

Keywords

• steganography
• digital security
• software agents lsb steganography
• ensemble classifier
• steganography datasets