14. Blind Quantitative Steganalysis Using CNN–Long Short-Term Memory Architecture

Covered communication is termed as steganography. And unveiling details about this camouflaged communication is steganalysis. In the current era, steganalysis helps forensics examiners to analyze and detect malicious communication. Feature extraction, training phase and testing phase are three steps used by traditional machine learning algorithms for learning complex characteristics of images before using them for analysis. Deep learning models, on the other hand, comprise a large number of hidden network layers which are used to capture various statistical information of digital images and learn them automatically. This paper proposes a novel deep neural network technique using convolutional neural network (CNN) with long short-term memory (LSTM) for quantitative steganalysis, prediction of embedded message length in given stego-object. Experiments are performed using Keras Python library. Statistical measure MAE is used to compare proposed quantitative steganalysis. Experimental results show the good performance of the proposed model.