.png){kind=icon}
.png){kind=icon}
Return
Reversible image hiding algorithm based on compressive sensing and deep learning
.png){kind=icon}
PRE
AI
DOI：10.1007/s11071-023-08516-5.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Compressive sensing (CS) can realize compression and encryption simultaneously. However, the current sampling-reconstruction algorithms based on CS are time-consuming due to the usage of large measurement matrix. Besides, CS-based image encryption also suffers from the large measurement matrix transmission issue and the attacker can easily distinguish whether an image is encrypted. This paper first proposes a novel three-dimensional chaotic map to improve randomness of generated keystream. A new reversible image hiding scheme is then realized using semi-tensor product compressive sensing and deep learning for mitigating computation, transmission and security concerns. In addition, a new Sine Logistic chaotic map (SineLCM) is constructed with its performance evaluated by interval boundedness, Lyapunov exponent and NIST. Especially, SHA-256 is taken to compute the hash values of the secret plain image. By grouping, three plaintext messages are generated by combining six random numbers. Then, a novel model is built to convert them into the initial keys of SineLCM with ciphertext messages open, according to elliptic curve encryption algorithm. As a result, the content-related mechanism can be established with no extra transmission. After being processed, the cipher image is embedded into the single-digit of high frequency coefficients got from integer wavelet transform performed on the carrier image. Experiments demonstrate that the proposed algorithm can achieve good performances, especially in running speed and image reconstruction quality. For example, the reconstruction for an image with size 512 x 512 only needs 0.6723 s.
Keywords:
Semi-tensor product compressive sensing
Discrete chaotic map
Image encryption
Reversible image hiding
Deep learning
Journal
IF:
6
Papers:
1.4W
Citations:
4.1W