A robust image encryption method based on block diagonal chaotic architecture with CNN performance evaluation metric

作     者：Mamandi, Vida Ardalani, Nahid Ghalamkari, Behbod 

作者机构：Department of Electrical Engineering Central Tehran Branch Islamic Azad University Tehran Iran Department of Mechanical Electrical and Computer Engineering Science and Research Branch Islamic Azad University Tehran Iran 

出 版 物：《Multimedia Tools and Applications》 (Multimedia Tools Appl)

年 卷 期：2025年

页      面：1-35页

核心收录：

学科分类：0810[工学-信息与通信工程] 08[工学] 0835[工学-软件工程] 0701[理学-数学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：Fading channels 

摘      要：In this paper, a new block diagonal chaotic model (BDC) is investigated due to higher necessity of advanced secure data transmission method in wireless medium and considerable limitation on computational storage space. In this model, the image sparse coding using BDC matrix in noisy and Rayleigh fading channels is studied. The structured BDC model is designed with two schemes, including chaotic pseudo inverse matrix (CPI-m) and block diagonal chaotic structured matrix (BDC-m) using two chaotic maps, in which the initial values for generating nonlinear chaotic sequences are used as private key at the receiver side. The simulation results show that the retrieved data measured based on CPI and BDC schemes outperforms the chaotic compressive sensing (CCS) method over AWGN and Rayleigh fading channels by 6% and 22.7%, respectively. Also, the sensitivity of recovered BDC-based image increases to 10(−32) compared to that of 10(−16) for conventional CCS based image. Moreover, the encryption key space increases from 10(37) for standard CCS-based measurements to β×10111 for measured data based on the BDC model, which guarantees that the competitors access no information. Considering datasets from iterative reconstruction process, the image features are extracted using convolutional neural network (CNN) model by ResNet-18 and GoogleNet architectures. The improved performance of data based on CPI and BDC model is validated by retrieved images accuracy over the classifier support vector machine (SVM). © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.