A Faster and Robust Artificial Neural Network Based Image Encryption Technique With Improved SSIM

Abstract

A robust image encryption process is still one of the most challenging tasks in image security owing to massive degree and sensitivity nature of information in the form of pixels. The hurdles include greater computational difficulty, information loss during encryption, universality, applicability of the approach, and less scalability. Many image encryption methods existing in literature merely encrypt a portion of the data. Therefore, we propose a robust, dynamic, and sophisticated technique to enhance the encryption process to make it difficult for an attacker to gain unauthorized access to the pixel data. The proposed system uses a novel analytical research methodology through dynamically harnessing the potential of neural network that offers better forward and backward secrecy, dynamic control, and automatic management unlike any existing system. The encryption procedure comprises of two levels, first level is confusion- permutation of input image and second level is diffusion by Bit XOR operation for secure transmission and storage of images. Finally, the encrypted image is used as a target for training the Artificial Neural Network (ANN) model. ANN trained values are used for final level of encryption to develop a Neural Network (NN)-based cryptosystem, where the crypto analyst or the cracker need to know the number of adaptive iterations and the final weights for the encryption and decryption systems to crack the system which offers higher degree of resiliency towards potential threats. Results and security analysis show that our algorithm has good encryption effect, ability of resisting exhaustive attack, statistical attack, and differential attack. The system performance after implementing the proposed method is compared with existing methods present in literature with respect to processing time and Structural Similarity Index Measure (SSIM). Our proposed method offers significant reduction in encryption time and is approximately 10-15% faster than others with SSIM of 0.002165, close to zero after encryption. It also successfully balances the image quality with higher image security and lower computational complexity.

Keywords

• Artificial neural network
• encryption
• image security
• SSIM
• NPCR
• UACI