Implementation and Empirical Evaluation of Pixel Value Differencing (PVD) Steganography with Boundary Mitigation Techniques

Abstract

This paper presents the idea of a cascaded boundary mitigation scheme for Pixel Value Differencing (PVD) steganography by jointly using modulus operation and MSB preserving bit-stream adjustment to address the falling-off-boundary issue. While the behavior of the applications of modulus and bits stream adjustment has been studied separately, their interaction and their performance at large scale have never been studied together. The method was implemented in Python and tested on the full dataset of Flickr30K (31,783 real world photographs) and a fixed payload of ~0.95 bits/pixel. Across all the images the framework achieved an average PSNR of 40.2dB (with a range of 34.3-45.1 dB), perfect-block rate (successful embedding without fallback) of 94.1% (with a range of 74%-100%), and extraction accuracy of 98.2% (100% on 28,147 images). The average capacity loss due to the fallback mechanism was 6.4%. Detailed results on five representative images of different texture are provided which confirm the robustness of the method for smooth as well as highly textured content. The approach preserves the linear O(M×N) complexity and provides a reproducible baseline to compare the traditional and the AI enhanced steganography techniques.