Abstract
Most of the computational time in many chaotic image encryption schemes is spent generating the required chaotic sequences. Since chaotic systems are defined by recurrence relations, they are often generated sequentially. In this paper, we propose a chaotic image encryption scheme which enables pixel-level parallelism to boost the computational speed of generating chaotic sequences. We use a group defined over elliptic curve (EC) points and the addition operator to generate a discrete chaotic sequence and use it to construct an image encryption scheme. The proposed scheme is designed such that encryption and decryption operations are highly parallelizable to take advantage of readily available parallel processing platforms such as GPU acceleration, DSPs and multi-core CPUs. Complexity analysis indicates that the proposed scheme is more efficient than existing EC-based image encryption schemes. Practical experiments on a quad-core CPU show that the proposed scheme can achieve a speedup of 3.93, confirming its superior parallelization efficiency in comparison with existing parallel image encryption schemes. We also provide detailed analysis of the immunity of the proposed scheme to all common cryptanalysis attacks. Results reveal that the proposed technique shows promising performance in terms of security and efficiency.