Detecting Hidden Data in Images Using Deep Learning

Data cloaking and digital concealment techniques pose a growing security challenge in modern communication environments, where the spatial dimension of images is exploited to transmit secret messages or malicious data. In contrast, steganography techniques are designed to detect these hidden modifications. This paper proposes a model based on deep learning, specifically convolutional neural networks (CNNs), supported by Spatial Rich Models (SRM) filters, to detect LSB data embedding. We review the latest studies on the development of cloaking and detection techniques and then present our experimental model, which was evaluated on a locally constructed dataset. The results demonstrate the model's ability to extract accurate statistical features and achieve a 91% recall rate for cloaked images. Data security technologies require the development of faster and more automated systems. This necessitates the development of a fast and automated cloaking analysis system that operates differently from traditional manual verification methods. Deep learning-based data hiding analysis systems employ a convoluted neural network (CNN) that classifies images into easy-to-lead and hidden images. Easy-to-lead images do not contain hidden data, while hidden images do. The methodology relies on three main phases: Data preparation: Data hiding analysis is performed using digitally combined random text images modified using Least Significant Bit (LSB) technology. The generated images form a balanced dataset appropriately classified for model training. Model building and training: The model is built and trained in layers. Each layer is responsible for amplifying the minor statistical variations resulting from the data hiding analysis process. The model is then trained on the dataset, attempting to associate underlying features that may not be visible to the naked eye. Evaluation and analysis: The model is evaluated based on criteria of accuracy, precision, and recall. The model has been shown to perform excellently in image classification, as a result of the weighted criteria. This also demonstrates the impact of deep learning in the practice of data hiding. The results of this research confirm that artificial intelligence techniques improve data security systems, and also confirm that deep learning models can be a powerful and effective tool for identifying taught activities in media content.