No embedding: A novel image cryptosystem for meaningful encryption.

We present a non-embedding image cryptosystem to accomplish the meaningful encryption. Thanks to the non-embedding superiority, the encrypted images are robust to both human eyes and steganalysis methods.We incorporate coupled dictionary learning and compressive sensing techniques into an unified framework, and further apply it to accomplish the task of non-embedding image encryption.Our scheme can reconstruct an image of more information with the image of little information. In this paper, we propose a novel image cryptosystem, which enables to encrypt the secret images with a smaller-size cover image. Compared with the existing meaningful encryption methods, our cryptosystem has three advantages: (1) non-embedding encryption, i.e., there isnt any data embedding into the cover image during the encryption process. (2) Our cryptosystem can simultaneously encrypt multiple secret images with one cover image, which greatly improves the security of secret images. (3) Our cryptosystem can accomplish not only the meaningful encryption, but also the meaningless encryption. Thus, people dont switch encryption methods when meeting different encryption requirements. Our scheme leverages the popular coupled dictionary learning and compressive sensing techniques to accomplish the whole task. Specifically, we use the coupled dictionaries to build connection between the cover image and the secret image, and apply the compressive sensing to decrypt the secret image. To demonstrate the effectiveness of the proposed cryptosystem, a series of experiments are conducted. Experimental results on gray images and colorful RGB images verify its superiority.