IAAE-Stega: Generic Blockchain-Based Steganography Framework via Invertible Adversarial Autoencoder

Abstract

Steganography is used to transmit secret messages over public networks, which is widely used in sensitive data transmission, anti-censorship systems, etc. Traditional steganography mainly embeds information into texts, images, and videos, but it is susceptible to tampering and tracking. Blockchain has the characteristics of anonymity, non-tampering, and flooding, making the blockchain-based steganography promising for secret messaging. However, existing schemes mainly focus on the generation of message-embedded fields and overlook the impact of required extra fields on concealment. Research results show that required extra fields can greatly increase the detection rate of transactions, up to 30%. Meanwhile, the embedding rate of blockchain-based steganography is low. If information can be embedded in these fields, the transmission capability of blockchain-based steganography can be improved. Current schemes for generating these fields face challenges such as low embedding rate, low concealment, and low efficiency. We propose an invertible adversarial autoencoder (IAAE) model. Different from ordinary AAE, IAAE consists of an invertible architecture, such as 1×1 convolution and fully connected layer, to ensure the information recovery ability. Based on IAAE, we propose IAAE-Stega, which uses IAAE to generate required extra fields. IAAE-Stega is able to embed information in required extra fields and make them indistinguishable from normal fields. In IAAE-Stega, the encoder is employed to hide information and generate indistinguishable required extra fields. After receiving a set of required extra fields, the decoder is employed to extract information. Experiments show that IAAE-Stega is better than all schemes in baselines and achieves state-of-the-art performance.