Deep Learning-Based Image Steganography Detection Using AI-Powered Systems

Abstract

Because it hides important information in seemingly commonplace media like images, audio files, and video files, steganography is crucial to secure digital communication. This method guarantees data confidentiality and shields sensitive data from unauthorised access or transmission interception. However, conventional detection methods like least significant bit (LSB) analysis, chi-square testing, and visual inspection have become less successful due to the quick development of steganography embedding algorithms, many of which incorporate adaptive and deep learning-based techniques. To provide good accuracy and generalizability for steganography detection, the existing study employs an intelligent steganography framework that is a hybrid mixture of machine learning and artificial intelligence. The proposed system makes use of a hybrid combination of deep techniques architecture with convolution neural networks (CNN), followed by feature extraction methods such as entropy calculations, pixel relation analysis, and histogram-based image texture evaluation. The CNN model learns more of special representation for more accurate identification of hidden information from the images. The extracted features will capture the low-level anomalies from the image structure. Evaluation was done based on the datasets – BossBase, COCO & image net which contain both normal and stego images. The results conclude that the CNN model performs the best approach rather than the traditional methods in terms of performance metrics such as recall, precision, accuracy, etc. Hence, with the power of a deep AI and CNN combination, the proposed methodology suggested a significant enhancement in the steganalysis.