DPM-EC²: A Distribution-Preserving ML-Guided Hybrid DWT–DCT Framework for Secure Image Steganography

Abstract

Image steganography has raised as a potent technique for the reliable transmission of records by concealing confidential information within virtual media while preserving visual imperceptibility. Nevertheless, many current approaches are susceptible to statistical steganalysis, exhibit limited adaptability to photographic characteristics, and employ fixed embedding strategies. This study presents a hybrid transform-area steganographic framework named DPM-EC2 to address these issues. To exploit multi-decision decomposition and electricity compression for eco friendly data embedding, the proposed model integrates the two-level Haar discrete wavelet (DWT) with block-based cosine transform (DCT). A portable device utilizing a learning-based approach is employed to estimate embedding costs and identify statistically secure regions for adaptive embedding. The payload is preprocessed through logistic chaotic randomization, Arnold scrambling, and BCH error correction coding to enhance security and robustness. Furthermore, Syndrome Trellis Coding (STC) in conjunction with a histogram restoration mechanism is used to preserve statistical distributions and minimize embedding distortion. Experimental evaluations on widely used benchmark images demonstrate excessive imperceptibility, with peak values of signal-to-noise ratio (PSNR) around 51 dB and values of structural similarity index measures (SSIM) near 0.999, indicating minimal visible distortion. The robustness assessment confirms the efficacy of the proposed framework. Overall, for reliable photo-based communication, the suggested method effectively balances imperceptibility, robustness, and resistance to steganalysis.