Researchers Keyvan Barati, Abdolsamad Hamidi, and Ali Farmani have developed a unified privacy-aware visual intelligence framework designed to secure sensitive visual data while enabling machine interpretation. Detailed in Scientific Reports, the computational architecture addresses the growing demand in intelligent visual systems to safeguard plaintext imagery without compromising downstream classification tasks. The system pairs chaotic image encryption with a learning pipeline that incorporates a classical nonlinear feature mapping inspired by quantum feature-space representations.
To obscure visual data, the encryption phase applies pseudo-random key generation alongside permutation–diffusion dynamics. The process introduces dual-key XOR transformations specifically structured to disrupt the spatial layout of original plaintext images. According to statistical security assessments conducted by the authors, this encryption stage demonstrates pronounced randomness and strong differential-sensitivity properties. The resulting ciphertexts yielded a mean number of pixels change rate of 99.8709 percent and a mean unified average changing intensity of 33.4629 percent, accompanied by a mean ciphertext entropy reaching 7.9972 bits.
Alongside the security mechanisms, the framework implements a learning branch that unites deep features with handcrafted visual representations. These extracted hybrid characteristics are subsequently transformed through the quantum-inspired nonlinear mapping prior to the classification stage. The investigators evaluated five distinct machine-learning models, selecting an architecture designated as Proposed-QINNFM-SVM. When evaluated on an independent held-out testing dataset, this selected model configuration achieved 100.0000 percent accuracy, a Macro-F1 score of 1.0000, and a balanced accuracy of 100.0000 percent.
To differentiate independent-test performance from the variability across training-data splits, the researchers also conducted repeated stratified five-fold cross-validation on the training data. Across the five evaluated configurations, the Proposed-QINNFM-SVM architecture demonstrated the highest stability and performance among the tested candidates. Under this cross-validation protocol, the model achieved a mean accuracy of 91.03 ± 0.20 percent, a Macro-F1 score of 0.8894 ± 0.0026, and a balanced accuracy of 87.10 ± 0.24 percent.
The authors note that separating the independent held-out evaluation from cross-validation partitions clarifies both baseline model capability and partition-dependent variance. Overall, the integrated computational architecture demonstrates a dual-purpose strategy for image protection and automated classification. The research team highlights potential future utility in sensitive visual-analysis settings, including medical imaging applications where data privacy and analytical reliability must be maintained simultaneously.
Subject of Research: Privacy-aware image encryption and machine learning classification
Article Title: Integrating chaotic encryption with a quantum-inspired nonlinear feature mapping for privacy-aware image intelligence
Article References: Barati, K., Hamidi, A., & Farmani, A. (2026). Integrating chaotic encryption with a quantum-inspired nonlinear feature mapping for privacy-aware image intelligence. Scientific Reports. https://doi.org/10.1038/s41598-026-80946-8
Image Credits: AI Generated
DOI: 10.1038/s41598-026-80946-8
Keywords: chaotic image encryption, quantum-inspired feature mapping, privacy-aware visual intelligence, permutation-diffusion dynamics, dual-key XOR transformation, pseudo-random key generation, ciphertext entropy, NPCR, UACI, hybrid feature extraction, support vector machine, Proposed-QINNFM-SVM, stratified cross-validation, medical imaging security, differential sensitivity, image classification
Cite Scienmag News
Katie Riggs. (November 22, 2026). Chaotic Encryption and Quantum-Inspired Mapping Protect Sensitive Images During Machine Interpretation. Scienmag. https://scienmag.com/chaotic-encryption-and-quantum-inspired-mapping-protect-sensitive-images-during-machine-interpretation/
Katie Riggs. "Chaotic Encryption and Quantum-Inspired Mapping Protect Sensitive Images During Machine Interpretation." Scienmag, 22 November 2026, https://scienmag.com/chaotic-encryption-and-quantum-inspired-mapping-protect-sensitive-images-during-machine-interpretation/. Accessed 22 November 2026.
Katie Riggs. "Chaotic Encryption and Quantum-Inspired Mapping Protect Sensitive Images During Machine Interpretation." Scienmag. November 22, 2026. https://scienmag.com/chaotic-encryption-and-quantum-inspired-mapping-protect-sensitive-images-during-machine-interpretation/








