Steganalysis for Still Images with LSB Steganography Using Machine Learning Algorithms

This paper investigated the application of machine learning for steganalysis using a feature-based dataset extracted from still images with the Least Significant Bit (LSB) steganography. We evaluated several models and found that Artificial Neural Networks (ANNs) achieve the highest classification accuracy within practical training times. The accuracy, however, is limited to 93 % due to constraints within the dataset. To overcome this barrier, more comprehensive datasets and/or models should be examined in future.