Secure Data Exchange: An Integrated Approach Using Convolutional Neural Networks and Steganography on MNIST Digital Number Images

Abstract

With the widespread use of digital communication, secure data transfer has become a top priority nowadays, which is important to prevent data leaking and hacking. This paper introduces a new concept that combines steganography and deep learning to protect confidential data during transfers. It uses the Modified National Institute of Standards and Technology (MNIST) dataset of handwritten digits to create a steganographic system that hides the required text within the images; before the steganography process, the text is encrypted with a cryptography technique, and for better security, a new approach of securing text is used, which make the encrypted text undetectable, and this technique adds additional level security, especially when it is merged with steganography. Also, this steganographic technique is integrated with a Convolutional Neural Network (CNN), resulting in a dual-purpose model that both disguises and predicts the presence of hidden messages with an impressive 99% accuracy. This method ensures that only individuals with the correct decryption key can access the confidential information, thereby ensuring the security of the transferred data. The fusion of these two methods sets a new standard for secure data exchange in modern networks, offering a robust framework that mitigates the vulnerabilities of current communication systems. Consequently, the detection of image-based steganography is highly accurate and significantly contributes to the development of data protection in the face of emerging artificial threats.