<p>Security of visual data between edge and cloud platforms necessitates algorithms that not only provide diffusion and confusion effectively but also have a viable computational cost. In this paper, we propose a near-lossless hybrid image encryption system that integrates transform-domain and spatial-domain processing by channel-collapsed operations. The multi-channel images are merged into single matrices to process them more efficiently, while the reversible local mixing by means of a single-level integer lifting Haar transform and the targeted LL-subband permutation is used to break frequency correlations. The transform coefficients are subjected to precision-controlled quantisation to an 8-bit dynamic range in order to reduce the storage space required while at the same time preserving the visual quality. Spatial randomisation is achieved with a mask-hardened nonlinear diffusion applied in two passes. The keystream is generated from two chaotic sources: a 5D hyperchaotic logistic-sine system and a 3D Lorenz attractor, combined with BLAKE3 keyed expansion. This setup provides high keystream diversity, unpredictability, and integrity. Security tests on standard images confirm the strength of the scheme: NPCR <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\varvec{\approx 99.61\%}\)</EquationSource> </InlineEquation>, UACI <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\varvec{\approx 35.28\%}\)</EquationSource> </InlineEquation>, entropy close to the maximum (7.997 bits), and correlation absolute values below 0.002. The framework is able to deliver a throughput of <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\varvec{\approx 2.65}\)</EquationSource> </InlineEquation> MB/s with a very small reconstruction error (PSNR <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\varvec{\approx 48}\)</EquationSource> </InlineEquation> dB), which makes it suitable for scenarios where secure storage is needed with a quality trade-off. Key-sensitivity experiments reveal that the avalanche effect is complete, i.e., one-bit changes in the key result in <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(\varvec{\approx 99.61\%}\)</EquationSource> </InlineEquation> pixel diffusion. The full code and the scripts that reproduce the results of the proposed method are available at: <a href="https://doi.org/10.5281/zenodo.18478414">https://doi.org/10.5281/zenodo.18478414</a>.</p>

A robust hybrid image encryption framework based on transform and spatial domain processing with high-dimensional chaotic maps

期刊论文
发表日期：

2026年2月13日

查看全文

Rafik Hamza

摘要

Security of visual data between edge and cloud platforms necessitates algorithms that not only provide diffusion and confusion effectively but also have a viable computational cost. In this paper, we propose a near-lossless hybrid image encryption system that integrates transform-domain and spatial-domain processing by channel-collapsed operations. The multi-channel images are merged into single matrices to process them more efficiently, while the reversible local mixing by means of a single-level integer lifting Haar transform and the targeted LL-subband permutation is used to break frequency correlations. The transform coefficients are subjected to precision-controlled quantisation to an 8-bit dynamic range in order to reduce the storage space required while at the same time preserving the visual quality. Spatial randomisation is achieved with a mask-hardened nonlinear diffusion applied in two passes. The keystream is generated from two chaotic sources: a 5D hyperchaotic logistic-sine system and a 3D Lorenz attractor, combined with BLAKE3 keyed expansion. This setup provides high keystream diversity, unpredictability, and integrity. Security tests on standard images confirm the strength of the scheme: NPCR \(\varvec{\approx 99.61\%}\) , UACI \(\varvec{\approx 35.28\%}\) , entropy close to the maximum (7.997 bits), and correlation absolute values below 0.002. The framework is able to deliver a throughput of \(\varvec{\approx 2.65}\) MB/s with a very small reconstruction error (PSNR \(\varvec{\approx 48}\) dB), which makes it suitable for scenarios where secure storage is needed with a quality trade-off. Key-sensitivity experiments reveal that the avalanche effect is complete, i.e., one-bit changes in the key result in \(\varvec{\approx 99.61\%}\) pixel diffusion. The full code and the scripts that reproduce the results of the proposed method are available at: https://doi.org/10.5281/zenodo.18478414.