Payload-Independent Direct Cost Learning for Image Steganography

Domain Transformation of Distortion Costs for Efficient JPEG Steganography with Symmetric Embedding

“…where X, Y and m represent the cover image, stego image and the secret messages, respectively. Emb(•) denotes the embedding conducted by the steganographic codes, D(•) represents the distortion of transferring the cover into a stego image, and H refers to the parity-check matrix of code C, while C(m) is the coset corresponding to syndrome m. At present, various steganographic schemes have been proposed for designing distortion cost functions with symmetric embedding, which can be mainly categorized into three groups: (1) heuristically designed schemes, such as WOWs (Wavelet-Obtained Weights) [10], UNIWARD (Universal Wavelet Relative Distortion) [11], HiLL (High-pass, Low-pass, and Low-pass) [12], QMP (Quaternion Magnitude-Phase) [13], UERD (Uniform Embedding Revisited Distortion) [14] and GUED (Generalized Uniform Embedding Distortion) [15]; (2) statistical-model-based schemes, such as MG (Multivariate Gaussian) [16], MiPOD (Minimizing the Power of the Optimal Detector) [17] and JMiPOD (JPEG steganography by MiPOD) [18]; and (3) deep-learning-based schemes, such as UT-GAN (U-net and double-Tanh framework using Generative Adversarial Network) [19], SPAR-RL (Steganographic Pixel-wise Actions and Rewards with Reinforcement Learning) [20], PICO-RL (Payload-Independent Cost learning framework using RL) [21], JS-GAN (JPEG Steganography using a GAN) [22] and JEC-RL (JPEG Embedding Cost with RL) [23]. It is worth noting that each of these designs can only be implemented on a specific type of image format, either spatial or JPEG; e.g., HiLL is specifically designed for spatial images and cannot be used for JPEG images.…”