High-Capacity Coverless Information Hiding Based On Robust and Deep Learning

Abstract

The rapid growth of multimedia communication over public networks has increased the need for secure and reliable information-hiding techniques. Conventional steganography generally embeds secret information by modifying pixels or other carrier components, which can introduce detectable statistical traces and create an attack surface for steganalysis [1]. Coverless Information Hiding (CIH) provides an alternative approach in which the carrier image is not directly modified. Instead, intrinsic image characteristics are extracted and associated with secret information through a mapping mechanism [2–4]. This paper proposes a high-capacity coverless information-hiding framework based on Local Binary Pattern (LBP) hashing, overlapping image blocks, lookup-table mapping, and a Wasserstein Generative Adversarial Network (WGAN). LBP is a computationally efficient texture descriptor originally developed for gray-scale and rotation-invariant texture analysis [5]. WGAN introduces a Wasserstein-distance-based objective intended to improve the stability of adversarial training compared with conventional GAN formulations [6]. The proposed method divides an image into blocks and sub-blocks, extracts LBP-based features, generates 8-bit hash codes, and maps secret information to corresponding image locations. A WGAN-based bidirectional transformation is then employed to generate an independent natural-looking carrier for secure communication. Experimental results reported in the study demonstrate up to 491,520+ bits/image embedding capacity using a single image. The reported recovery rate is 100%, with PSNR = ∞ dB and SSIM = 1.000 under the stated experimental conditions. Robustness experiments include JPEG compression, cropping, occlusion, geometric transformations, noise, filtering, and histogram equalization. The proposed framework therefore provides a high-capacity, database-independent approach to coverless information hiding with an emphasis on robustness and reversible secret-data recovery.