PA-SIE: Pixel-Attack Methodology Factored Selective Image Encryption Scheme for Low Resource Edge Computing Systems

The transmission of image data over the internet is inevitable considering its effective use-cases in tele-healthcare, online-banking, remote-education, e-commerce, email exchanges, and many others. However secure communication of the image data is still evolving and not standardized yet, but it is desirable owing to various cyber attacks especially adversarial attack leading to the loss of critical information. One of the primary challenges on adopting secure communication is to balance computational and communication cost for encrypting the data but also attain high security. Applying encryption across all pixelated data is likely to increase the communication overhead and reduce the throughput which otherwise becomes bottleneck for some of real-time applications. The large size of ciphertext makes the system highly inefficient. This research proposes an approach based on one-pixel attack factored selective image encryption scheme (PA-SIE) to reduce the size of the ciphertext while ensuring the security of the image data. The proposed approach identifies the sensitive pixels in the image by using the differential evolution algorithm and encrypts them using a preferred encryption scheme, while leaving the remaining pixels unencrypted. To the best of our knowledge, this is the first study to propose this approach for balancing cost and security in image data transmission. This approach significantly reduces the cost of data transfer while ensuring the secure transmission of sensitive image data. The proposed PA-SIE approach offers maximum communication latency improvement and energy savings of 84.65 %, when compared with full Elliptic Curve Cryptography (ECC) applied encrypted image transmission for web traffic over the HTTPS protocol. The proposed approach is a step towards ensuring secure and efficient transmission of image data over the internet from low resource and bandwidth constrained edge-computing systems.