A visually meaningful image encryption scheme based on a 5D chaotic map and deep learning

Abstract

ABSTRACT The VMIE techniques follow two steps: pre-encryption and embedding. In the first phase of the proposed method, the 5D chaotic map and zig-zag transformation are used for encryption, which provides higher security. In the second phase, the U-Net architecture-based encoder is implemented to hide the encrypted image in the reference image. Furthermore, an efficient decoder is designed to extract the encrypted image from the reference image. The exquisiteness of deep learning in image embedding is that it hides an image into the same size reference image without degrading the image quality. Moreover, to validate the proposed method, some standard security and image quality-related parameters are analysed. The results in terms of image security and quality parameters compared to the existing VMIE reveal that the proposed method is highly secure and efficient for imperceptible image communication and better image quality at the receiver’s end.