A Coverless Image Steganography Based on a Robust Object Detection Network

Compared with traditional image steganography algorithms, coverless image steganography has attracted a lot of researchers due to the ability to resist steganalysis algorithms completely. However, most existing methods have yet to be designed specifically to resist geometric attacks. The robustness against geometric attacks could be better. In this paper, a novel coverless image steganography based on a robust object detection network is proposed. First, a mapping dictionary is established for both senders and receivers. Then, a robust object detection network is constructed to detect objects in all images. This object detection network contains two specific-designed operations, denoted as Add Noise and Add Confidence, which improve the robustness by introducing the Gauss noise and the confidence of labels. After that, the index structure is designed to accelerate the speed of searching. Finally, the secret information is divided into fixed-length segments and mapped to the stego-images according to the mapping dictionary to accomplish information hiding and extraction. Experimental results show that the proposed method performs much better in robustness than the state-of-the-art methods.