High-capacity image steganography secured by universal adversarial perturbations

Steganography embeds a significant amount of information within various data types, such as text, images, audio, or video, in a manner that complicates the presence of the hidden material or any detection efforts. The presented method improved the data-hiding capability and Security by employing Discrete Cosine Transform, Universal Adversarial Perturbation (UAP), and Discrete Wavelet Transform. A two-stage steganographic scheme has been developed in which secret data are first hidden within selected DCT coefficients of a two-level DWT decomposition and then subjected to a constrained Universal Adversarial Perturbation. UAP perturbs the cover image in a way that resists deep learning attacks. A scaling factor is calculated using the 2D DWT transform of the cover image and the DCT transform of the message image. By uniting high-capacity transform-domain embedding with adversarial noise generation, the approach achieves both substantial payload density and robust security against deep-learning-based detectors.