Deep Learning-Based Steganography: A Neural Network Approach for Secure Data Embedding

Abstract

The urgent need for secure communications in the digital era has led to the development of steganography algorithms. Steganography is a technique for hiding digital data in a media such as an image. Traditional techniques, despite their effectiveness, face some limitations such as imperceptibility, payload capacity, and robustness against attack. The emergence of Artificial Intelligence (AI), especially Deep Learning (DL), has revolutionised the field of data hiding and security. This study reinforced recent developments by highlighting deep learning models by training Convolutional Neural Networks (CNNs) and extracting features such as edges, boundaries, and color contrast between each pixel and its neighbours. When the pixel is selected, it is embedded in the Least Significant Bit (LSB), thus obtaining a high payload capacity of the data, imperceptibility, and a more robust image against attacks. The results obtained proved the worth of the proposed method, as BSNR = 92 dB and MSE = 25 were obtained to produce an image with a higher payload capacity and more security. In the future, the integration of algorithms such as machine learning and deep learning can be utilized to create a hybrid algorithm that is better in terms of statistical attacks