Image Steganalysis based on Pretrained Convolutional Neural Networks

the process of identifying the presence of secret information in cover images is known as image steganalysis. As a result, classifying an image as a cover image or a stego image might be considered a classification task. The majority of steganalysis approaches that rely on deep learning are effective. Deep learning technology can identify and extract features mechanically using deep networks, allowing steganalysis technology to eliminate the need for specialist knowledge. However, Deep learning model training is tough and takes a large amount of processing time and information. Therefore, pre-trained CNN such as AlexNet model were used as feature extractors to save time during training. Therefore, this research presented an image steganalysis method based on AlexNet CNN Model. There are 3 steps make up the proposed image steganalysis method: Firstly, Data collection and preparation. Secondly, AlexNet model are used for extract Distinctive features. Lastly, the feature vector is then utilized to train the Random forest (RF) classifier in order to detect the binary classification (Cover/Stego). The experimental results under IStego100K database show that the proposed method accuracy is 99%. The properties of AlexNet models can be deduced to be useful and concise to classify using RF. In compared to previous techniques, the presented method outperformed them.