Visually meaningful image encryption based on universal embedding model

Visually meaningful image encryption (VMIE) means that a plain image can be encrypted into a visually meaningful cipher image (VMCI), which makes the secret more imperceptible than noise-like cipher images. Here, we first present a universal embedding model (UEM) and further present a new UEM-based VMIE algorithm. The plain image is pre encrypted and then embedded into the integer wavelet subbands of the host image in a dynamic way. In order to avoid the overflow, a threshold limiting function is used to modify the range of pixel values of the host image. To adapt different types of wavelet transform subbands, a UEM is proposed to be used to embed the secret information domain into the host domain. Moreover, in the embedding process, a four-dimensional discrete chaotic system is used to ensure the security of embedding. The numbers of embedded bits in the four embedded domains can be adjusted flexibly. A traversal algorithm is designed to find the optimal numbers of embedded bits for different types of wavelet transform in order to achieve an optimal visual quality of cipher images. A matrix-bit-depth conversion algorithm is designed where a large matrix with low bit depth can be transformed into a small matrix with high bit depth to meet the input requirements of UEM, which means that the size of the plain image is not limited to a quarter of the size of the host image. Simulation results and performance comparisons show that the proposed VMIE algorithm can achieve a better visual quality than existing VMIE algorithms. (c) 2021 Elsevier Inc. All rights reserved.

Automated summary

The study introduces a visually meaningful image encryption (VMIE) algorithm based on a universal embedding model, which achieves better visual quality by optimizing the number of embedded bits and flexibly adjusting wavelet transform subbands.