Adaptive Pixel Value Differencing with Machine Learning for Intelligent Steganography

This research explores the integration of Adaptive Pixel Value Differencing (APVD) with machine learning techniques to develop an intelligent, content-aware steganography system. The primary objective is to enhance data hiding efficiency, imperceptibility, and robustness by leveraging machine learning models to dynamically optimize the embedding process. The methodology involves training a Random Forest classifier to identify optimal pixel regions and embedding parameters based on local image characteristics, such as variance and texture. This model then guides the APVD algorithm for secret data insertion. Experimental evaluation on diverse image datasets (USC-SIPI and BOSSBase) assesses the system's performance in terms of Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Bit Error Rate (BER) under simulated noise and compression attacks. Key findings demonstrate that the machine learning-enhanced APVD method significantly outperforms traditional APVD, achieving an average PSNR increase of 2-4 dB and reducing the BER by up to 30% under attack conditions. This approach marks a significant advancement towards creating adaptive, intelligent steganographic techniques suited for secure communication in complex and dynamic digital environments.