An improved visually meaningful encrypted image scheme

An encrypted image generally presents an appearance of texture-like or noise-like image, which exhibits an indication of encrypted data that may motivate a significantly large number of attacks. To address this problem several approaches have been carried out, where an encrypted image is transformed into a visually meaningful encrypted image (VMEI). We present an improved approach to encrypt a plain image into a visually meaningful cover image, where the preprocessing and embedding phases are modified to achieve a quite comparable performance. To make the assessment of this proposal, some standard metrics such as PSNR and SSIM are performed, and the obtained numerical results show that this proposal is comparable or superior to some VMEI schemes used as benchmark in this work. In most of the reference images considered in the dataset the values of PSNR are greater than 40 dB, and a value close to 1 for the SSIM metric, which means a great similarity and high quality, respectively, between the reference images and their obtained VMEI images. Also, the two-dimensional detrended fluctuation analysis is carried out to observe the scaling behavior of the images, which allows us to have a perceptual security parameter of encrypted images and visually meaningful cover images. For this metric, the scaling exponent values of the VMEI images are close to 2.1, which are very similar to the values obtained for the reference images, a situation that confirms that the hidden information may not be detected by the human visual system. Furthermore, for two reference images, this metric was able to detect how the pixel values are distributed, and how the pixel values are highly correlated, suggesting that this metric can be an objective measure of encrypted images.