A Novel YOLOv2‐Guided Region of Interest Encryption Framework Based on Elliptic Curve, DNA Computing and Chaotic Dynamics
Su Lv, Abdul Rauf, Damia Tasleem, Adnan Aslam · IET Image Processing · 2026
ABSTRACT This study addresses the challenge of secure image transfer in an era of frequent data breaches by proposing a hybrid encryption technique focused on protecting regions of interest (ROI). The method integrates DNA‐inspired operations, elliptic curve cryptography (ECC) and elliptic curve Diffie–Hellma and chaotic systems to achieve high security. Initially, you only look once, version 2 is used to segment the image into ROI and non‐ROI areas. A user‐selected ROI is processed via secure hash algorithm 256 to generate a 256‐bit hash, which provides dynamic parameters for an ECC‐based key exchange. A novel 2D chaotic system, initialized using elliptic curve points, generates pixel coordinates for selective scrambling and substitution. A custom S‐box, derived from the hash and chaotic outputs, applies pixel‐level confusion using six dynamic coordinate rules. Encryption strength is further enhanced by incorporating DNA encoding for substitution and permutation. Experimental results confirm the schemes robustness, evidenced by near‐ideal entropy (7.999), high number of pixels change rate (99.61%), near‐zero correlation, uniform histogram and desirable low peak signal‐to‐noise ratio (8 dB) with high mean squared error (9000). The method demonstrates resilience against noise and shear attacks, with key sensitivity tests showing over pixel change from a single‐bit hash alteration. This ROI‐focused approach is ideal for applications requiring partial encryption, such as medical imaging, remote sensing and surveillance.