Saliency-Guided Enhancement and Hybrid 5D Chaotic-ECC Encryption for Secure Military Image Processing
TL;DR
A hybrid encryption technique involving 5D hyperchaotic key stream generation and elliptic curve cryptography is employed to provide high resilience against different cyber-attacks such as differential attacks, chosen-plaintext attacks, and side-channel attacks.
Abstract
In this paper, a novel two-stage framework for enhancing and securing the encryption of military images is presented. For stage one, a saliency-guided CNN autoencoder is employed to enhance visual quality by highlighting features of salient objects using Grad-CAM (Gradient-weighted Class Activation Mapping). Saliency guide mechanism would allow the network to dedicate representational capability to important tactical targets such as vehicles, radar, forts, etc while suppressing irrelevant background. For stage two, a hybrid encryption technique involving 5D hyperchaotic key stream generation and elliptic curve cryptography (ECC) is employed to provide high resilience against different cyber-attacks such as differential attacks, chosen-plaintext attacks, and side-channel attacks. We evaluated our framework with the MSTAR (Moving and Stationary Target Acquisition and Recognition) dataset and obtained results of 35.2 dB PSNR, 0.942 SSIM [9], 7.998 bits/pixel encryption entropy, 99.63% NPCR and 33.8% UACI. Our results significantly outperform the state-of-the-art competitor by 5.1 dB for PSNR and 0.091 for SSIM, and p<0.001 is obtained for all measures. The overall integrated pipeline takes 23.4 ms for a 256256 SAR image processing on GPU, which allows real-time implementation in UAV vision systems and battlefield communication security.