TPR-BBGAN: A Twister Pseudo-Random and Barzilai–Borwein Optimised Neural Cryptography Model for Secure Image Communication

The possibility of securing textual image data sharing exponentially strengthens when it harnesses the potential of cryptography as well as deep learning methods. A review of the existing literature showcases some interesting and productive initiatives; however, they are noted with issues, viz., increased reconstruction error, weak generation of pseudorandom keys, static threshold-based validation, etc. All these issues lead to suboptimal data integrity as well as confidentiality, which is a leading gap in research on neural optimised-based solutions. Therefore, the proposed system introduces an innovative Twister Pseudo Random and Barzilai–Borwein Gradient Autoencoder Neural Network (TPR-BBGAN) for secure textual image data sharing. The model introduces various novel operations, viz., feature extraction using fuzzy batch-normalised preprocessing, key extraction using the Barzilai–Borwein method, an autoencoder, and Mersenne Twister. The TPR-BBGAN determines the optimal threshold dynamically, contributing to a reduction in the reconstruction error while convergence performance is boosted. The experimental outcome shows that the TPR-BBGAN achieves a 12–20% enhancement in data confidentiality, a 6–17% enhancement in data integrity, a 30–46% reduction in bit-error rate, and a 6–20% increase in the Peak Signal-to-Noise Ratio (PSNR) in contrast to existing models.