ML-Based Sensitivity-Aware Selective Image Encryption Algorithm

With the rapid growth of image transmission in Internet of Things (IoT), edge computing, and multimedia applications, ensuring data confidentiality while maintaining low computational overhead has become a critical challenge. Traditional full-image encryption techniques provide strong security but are computationally expensive, making them unsuitable for resource-constrained environments. Selective image encryption (SE) addresses this issue by encrypting only sensitive regions; however, most existing approaches rely on static rules and fixed thresholds, limiting their adaptability and security robustness. This paper proposes a Machine Learning–Driven Adaptive Selective Image Encryption (ML-SEI) framework that intelligently identifies sensitive image regions and dynamically adjusts encryption strength. The proposed method integrates supervised machine learning models—Support Vector Machine (SVM), Random Forest (RF), Convolutional Neural Network (CNN), and Logistic Regression (LR)—to predict region sensitivity based on multi-dimensional image features. Among them, the SVM-based model is selected using a performance-aware strategy. Experimental results on public datasets demonstrate that the proposed ML-SEI achieves 96.4% sensitivity prediction accuracy , improves entropy and NPCR values, and reduces encryption time by 28–35% compared to traditional selective and full encryption schemes. The results confirm the suitability of the proposed framework for real-time, resource-constrained, and edge-based image security applications. Security robustness is evaluated under a ciphertext-only adversarial model using public benchmark datasets.