Enhancing Image Encryption for Efficient and Secure Transmission in Computer Networks: A Parameter-Optimized Approach for Embedded Devices

Our paper presents a novel approach to image encryption for secure transmission in computer networks, building upon the framework of "Convolution-based Image Encryption for computer networks." The existing model's size was found to be impractical, prompting us to implement significant modifications, employing Depthwise Separable Convolutions and Separable Transposed Convolution techniques. By doing so, we remarkably reduced the model parameters from ۷۵ million to approximately ۱۴ million, while still achieving a commendable PSNR score of ۳۲ for the decoder. Our asymmetric model combines adversarial training and image compression to effectively encrypt and compress image data. The encryption process involves transforming images into encrypted feature maps using convolutional layers, followed by lossy compression to reduce data size. Decryption reverses these steps to recover the original image. Through evaluation on standard image datasets, we establish the superiority of our proposed method over existing techniques in terms of both security and compression performance, promising a robust solution for secure image transmission in computer networks.

کلیدواژه ها:

Image compression ، image encryption ، deep learning ، adversarial training ، asymmetric model ، embedded devices