An Adaptive Security-Aware Image Encryption Model with Tamper Localization and Recovery Based on Compressive Sensing

This is the authors' abstract. We don't add key points for this paper.

Image protection has become a critical concern in the digital age, where images are constantly transmitted, stored and shared over various networks. However, traditional watermarking methods often suffer from low performance in terms of security, recovery capacity and robustness. Recently, Compressive Sensing (CS) techniques have been applied to solve these issues. It can offer a secure and computationally effective solution by leveraging data compression and image encryption as a single process. Thus, an innovative image authentication framework is developed in this work with tamper localization and recovery using CS and a watermarking mechanism. Initially, data is collected from benchmark databases and processed using the Non-subsampled Contourlet Transform (NSCT). High-frequency components are extracted and subjected to Arnold scrambling and CS using sparse representation, while low-frequency components undergo Arnold scrambling and watermarking. The processed low- and high-frequency images are then combined using Inverse NSCT and encrypted with Adaptive 2D Logistic Chaotic Encryption (A-2DLCE). Here, an Improved Single Candidate Optimizer (ISCO) is employed to select the optimal encryption key. During decryption, images are first processed by A-2DLCE and then split using NSCT into low- and high-frequency components. Watermark extraction is performed on low-frequency images, followed by inverse Arnold scrambling, tamper detection and recovery. High-frequency images are reconstructed using inverse Arnold sampling and CS, and all components are finally combined using Inverse NSCT to obtain the original images. The proposed framework is evaluated against existing methods to demonstrate its effectiveness. Here, the Mean Squared Error (MSE) of the developed model is 0.0153, the Number of Pixels Change Rate (NPCR) is 99.62% and the Unified Averaged Changed Intensity (UACI) is 33.71%. Thus, the results proved that the developed model decisively proves its enhanced performance by using the powerful combination of models. It also aids in dramatically improving the performance, even with complicated tasks, like anomaly detection and image reconstruction.