Quantum-Based Steganography on AI-Generated and Real Images: A Comprehensive Experimental Analysis and Security Evaluation

Abstract

Steganography provides an important mechanism for secure covert communication by concealing sensitive information within digital media. However, conventional steganographic approaches may be affected by predictable key-generation mechanisms, statistical detectability, and vulnerabilities to steganalysis. This paper presents a quantum-based steganographic framework that integrates quantum key generation with Least Significant Bit (LSB) image steganography and evaluates its performance on both real and AI-generated images. The quantum keys are created based on the quantum circuit design which uses the Hadamard transform. In particular, a single qubit initialized to |0⟩ is transformed into a uniform superposition using a Hadamard gate, followed by quantum measurement to generate unbiased binary key bits. The required (L)-bit quantum key is obtained by collecting (L) individual measurement outcomes using the Qiskit framework and Qiskit AerSimulator. Then, the keys are utilized for the embedding and extraction procedures. The proposed approach was tested on a dataset which consisted of 200 RGB images. There were 100 real images and 100 AI-generated images in the test set. First, a fixed payload of 2,296 bits was embedded into each image and evaluated using image quality metrics, extraction accuracy, embedding efficiency, steganalysis, and compression robustness. Scalability was then evaluated by increasing the payload size from 6,000 to 96,000 bits, representing a sixteen-fold increase in embedded data.

As a result of experiments, the images with the embedded data were found to be visually similar to the original images because their Peak Signal-to-Noise Ratio (PSNR) values are higher, Mean Squared Error (MSE) values are negligibly small and Structural Similarity Index (SSIM) values are almost equal to one. Therefore, it can be stated that the embedding introduces almost no distortion to the image and the proposed method performs equally well on real and AI-generated images. Moreover, a number of additional tests were performed to check how easily the embedded data could be detected and recovered from images processed by different means. The experimental results show that the proposed method achieves high visual fidelity, reliable hidden message recovery and consistent computational performance on both kinds of images. This study shows the possibilities of the integration of quantum key generation and image steganography for the purposes of secure covert communication and can be considered as a starting point for further investigations with physical quantum systems and more complex steganalysis techniques.