Faster-RCNN Based Robust Coverless Information Hiding System in Cloud Environment

Abstract

Key distribution is the foundation for protecting users' privacy and communication security in cloud environment. Information hiding is an effective manner to hide the transmission behavior of secret information such as keys, and thus it makes the secure key distribution possible. However, the traditional information hiding systems usually embed the secret information by modifying the carrier, which inevitably leaves modification traces on the carrier. Thus, they cannot resist the detection of the steganalysis algorithm effectively. To avoid this issue, the coverless information hiding technique has been proposed accordingly, in which the original images of which features can express the secret information are directly used as stego-images. Since the existing coverless information hiding methods use the low-level handcrafted image features to express secret information, it is hard for them to realize desirable robustness against common image attacks. Moreover, their hiding capacity is limited. To conquer these problems, we design a novel robust image coverless information hiding system using Faster Region-based Convolutional Neural Networks (Faster-RCNN). We employ Faster-RCNN to detect and locate objects in images and utilize the labels of these objects to express secret information. Since the original images without any modification are used as stego-images, the proposed method can effectively resist steganalysis and will not cause attackers' suspicion. The experimental results demonstrate that the proposed system has higher performance in terms of robustness and capacity compared to the typical coverless information hiding methods.

Keywords

• Coverless information hiding
• image steganography
• object detection
• faster-RCNN
• key distribution