A novel parallel image encryption algorithm based on hybrid chaotic maps with OpenCL implementation

Since chaotic maps have the excellent properties of unpredictability, ergodicity and sensitivity to their parameters and initial values, they are quite suitable for generating chaotic sequences for securing communication systems and are also especially useful for securing images, and a lot of chaotic map-based image encryption algorithms have been proposed. But some existing image encryption algorithms were proved that their security, encryption efficiency or computational speeds are not quite satisfactory for practical applications. Some of them using only one type of chaotic system may suffer from low key space, and some others using two or more types of chaotic system may suffer from high computational overheads. In this paper, based on the classic 1D logistic map, a 2D one-coupling logistic dynamics system and OpenCL, a novel parallel image encryption algorithm HCMO is proposed. Our algorithm consists of a confusion phase and a diffusion phase using four sub-key matrices based on the hybrid logistic dynamics systems, the linear transformation and the enlarging operation. In the confusion phase, the image's pixel positions are first scrambled by performing row-wise and column-wise permutation operations using two sub-key matrices; then, in its diffusion phase, both the bit XOR operation and the bit cyclic shifting are applied onto the scrambled intermediate image matrix using the other two sub-key matrices. In order to reduce the whole encrypting execution time, we speed up our HCMO on an OpenCL's heterogeneous and parallel characteristics. Compared to the implementation of Vihari's algorithm and some other chaotic map-based algorithms referred in this paper with the OpenCL-based implementation on the CPU and on the GPU, respectively, our algorithm's simulation demonstrates remarkable improvement in the operational speedup, and the experimental result analyses have also shown that HCMO has a higher-level security than some other referred algorithms.